An Analysis of Hydraulic, Environmental and Economic Impacts of Flood Polder Management at the Elbe River

Institute of Geoecology

An analysis of hydraulic, environmental and economic impacts of flood polder management at the Elbe River

A dissertation submitted to the Faculty of Mathematics and Natural Sciences at the University of Potsdam, Germany for the degree of Doctor of Natural Sciences (Dr. rer. nat.) in Geoecology

by Saskia Förster

Potsdam, August 2008

Online published at the Institutional Repository of the Potsdam University: http://opus.kobv.de/ubp/volltexte/2008/2726/ urn:nbn:de:kobv:517-opus-27260 [http://nbn-resolving.de/urn:nbn:de:kobv:517-opus-27260] Contents

Summary...... 7

Zusammenfassung ...... 9

Chapter I Introduction...... 11 1 Background ...... 11 2 Objective and Research questions ...... 12 3 Study areas...... 12 4 Outline of the thesis...... 14

Chapter II Flood risk reduction by the use of retention areas at the Elbe River...... 15 1 Introduction ...... 16 2 Description of the study area...... 16 3 Research approach ...... 17 4 Modelling the inundation of the polder system...... 17 4.1 Outline of the CR Model ...... 17 4.2 Application of the CR Model to the Lower Havel River ...... 18 4.3 Aggregation of simulation results ...... 20 5 Damage assessment in the polder system...... 20 6 Assessment of potential damage reduction in Wittenberge...... 21 6.1 Determination of flooded areas ...... 21 6.2 Assessment of potential damage in Wittenberge...... 23 7 Cost-benefit analysis ...... 23 8 Discussion and conclusions...... 24

Chapter III Hydrodynamic simulation of the operational management of a proposed flood emergency storage area at the Middle Elbe River...... 27 1 Introduction ...... 28 2 Study area...... 28 3 Hydrodynamic model setup...... 31 4 Considered flood scenarios and control strategies...... 32 4.1 Flood frequency analysis ...... 32 4.2 Definition of flood scenarios...... 33 4

4.3 Definition of control strategies ...... 33 5 Results...... 35 6 Conclusions and discussion ...... 37

Chapter IV Comparison of hydrodynamic models of different complexities to model floods with emergency storage areas...... 39 1 Introduction ...... 40 2 Study Area and Data Used...... 41 3 Methodology ...... 42 3.1 One-dimensional model setup ...... 42 3.1.1 Calibration and validation of model ...... 43 3.2 One-/ two-dimensional model setup...... 43 3.3 Sensitivity analysis ...... 44 3.3.1 Manning’s n...... 44 3.3.2 DEM’s of different resolutions ...... 44 3.3.3 Number of cross-sections used...... 44 3.3.4 Gate opening time and opening/closing duration...... 45 4 Results and Discussions ...... 45 4.1 Calibration and validation of the one-dimensional model ...... 45 4.2 One-dimensional model results for flooding and emptying processes in the polder...... 46 4.3 Comparison of one- and one-/ two-dimensional model results...... 48 4.4 Computation time, storage space requirements and modelling effort ...... 48 4.5 Sensitivity analysis ...... 50 4.5.1 Manning’s n...... 50 4.5.2 DEM’s of different resolutions ...... 51 4.5.3 Number of cross-sections used...... 53 4.5.4 Gate opening time and opening/closing duration...... 53 5 Conclusions...... 54

Chapter V Simulation of water quality in a flood detention area using models of different spatial discretisation ...... 57 1 Introduction ...... 58 2 Study site...... 58 3 Methods and data ...... 59 3.1 Comparison of the 0D and the 2D approach...... 59 3.2 Water quality processes ...... 60 3.3 Computational procedures...... 63 3.4 Boundary conditions, parameters, and initial values...... 63 3.5 Analysis of uncertainty ...... 64 4 Results...... 66 4.1 Simulation results of the two models...... 66 4.2 Uncertainty of predictions ...... 68 5

5 Discussion and conclusions...... 69 5.1 The impact of spatial discretisation ...... 69 5.2 Conclusions from the uncertainty analysis ...... 70 5.3 Implications and recommendations ...... 70

Chapter VI Assessing flood risk for a rural detention area ...... 73 1 Introduction ...... 74 1.1 Damage estimation methods ...... 74 1.2 Study site...... 76 2 Methodology ...... 76 2.1 Damage estimation...... 77 2.2 Flood frequency analysis ...... 79 3 Results...... 81 4 Sensitivity analysis ...... 83 5 Discussion and Conclusions...... 84

Chapter VII Conclusions and recommendations ...... 87 1 Conclusions...... 87 1.1 Hydrodynamic modelling...... 87 1.2 Water quality modelling...... 88 1.3 Vulnerability assessment...... 88 1.4 Model complexity...... 89 2 Recommendations...... 90 2.1 Flood polder management...... 90 2.2 Future research ...... 92

References ...... 95

Acknowledgements ...... 101

Curriculum Vitae and list of publications ...... 103

Author’s declaration ...... 105

SUMMARY

In recent years, extreme river floods have occurred in many European countries including Germany. These events have triggered a shift from traditional safety-oriented flood protection to a comprehensive flood risk management approach. Flood risk management comprises measures that aim at reducing the flood hazard as well as the vulnerability to floods. Flood polders are part of the flood risk management strategy for many lowland rivers. They are used for the controlled storage of flood water so as to lower peak discharges of large floods. Consequently, the flood hazard in adjacent and downstream river reaches is decreased in the case of flood polder utilisation. Flood polders are dry or partially wet storage reservoirs that are typically characterised by agricultural activities or other land use of low economic and ecological vulnerability. The objective of this thesis is to analyse hydraulic, environmental and economic impacts of the utilisation of flood polders in order to draw conclusions for their management. For this purpose, hydrodynamic and water quality modelling as well as an economic vulnerability assessment are employed in two study areas on the Middle Elbe River in Germany. One study area is an existing flood polder system on the tributary Havel, which was put into operation during the Elbe flood in summer 2002. The second study area is a planned flood polder, which is currently in the early planning stages. Accordingly, the studies can be addressed as ex-post and ex-ante investigations. Hydrodynamic numerical models are applied in both study areas to investigate peak reduction and flooding characteristics under different flood scenarios and operational schemes. It is demonstrated that flood peak levels can effectively be capped by temporary water storage. However, the research also shows that the obtainable flood peak reduction strongly depends on the shape of the flood hydrograph. Flooding characteristics such as water depth, flow velocity and inundation duration exhibit a large spatial variation throughout the flood polder areas. This information provides input for the subsequent water quality modelling and vulnerability assessment. Long water storage may have negative environmental effects. A major problem that has been observed in flood polders on the Elbe River during the August 2002 flood was the depletion of dissolved oxygen due to the strong oxygen demand imposed by organic material in the water body and on agricultural fields. Therefore, a numerical model is employed for the simulation of dissolved oxygen dynamics throughout the filling, stagnancy and emptying phases of the flood polder operation. The economic vulnerability assessment focuses on agricultural land according to the prevailing land use in the flood polders. In contrast to other damage categories, losses on agricultural fields exhibit a strong seasonal pattern, while the flooding probability also has a seasonal variation. Both seasonal aspects are included in the estimation of annual losses on agricultural land. Furthermore, numerical models of different spatial dimensionality, ranging from zero- to two- dimensional, are applied in order to evaluate their suitability for hydrodynamic and water quality simulations of flood polders in regard to performance and modelling effort. For many applications, models of lower dimensionality may deliver sufficiently accurate results, while providing the opportunity to investigate various scenarios and extensively test model sensitivity. In contrast, two-dimensional models require a considerably higher computational effort, but are inevitable for certain applications. Generally, two-dimensional approaches may be preferred in floodplain areas of complex topography and distinct spatial variations in hydraulic variables such as flow velocity and water depth. For practical applications, it is suggested to use a simplistic approach first and increase spatial dimensionality if the target variables show a large spatial variability. The thesis concludes with overall recommendations on the management of flood polders, including operational schemes and land use. In view of future changes in flood frequency and further increasing values of private and public assets in flood-prone areas, flood polders may be effective and flexible technical flood protection measures that contribute to a successful flood risk management for large lowland rivers. 8

ZUSAMMENFASSUNG

In den vergangenen Jahren traten in vielen europäischen Ländern einschließlich Deutschland extreme Fluss-Hochwasser auf. Diese Ereignisse haben einen Wandel vom traditionellen auf Sicherheit ausge- richteten Hochwasserschutz zu einem übergreifenden Hochwasserrisikomanagement angestoßen. Das Hochwasserrisikomanagement umfasst sowohl Maßnahmen zur Verminderung der Hochwassergefähr- dung als auch der Vulnerabilität gegenüber Hochwassern. Flutpolder sind Bestandteil des Hochwasserrisikomanagements an vielen Tieflandflüssen. Sie werden zum gesteuerten Rückhalt von Wasser eingesetzt, um Spitzenabflüsse von großen Hochwassern zu sen- ken. Dadurch wird im Falle des Flutpoldereinsatzes die Hochwassergefährdung für benachbarte und unterstrom liegende Flussabschnitte verringert. Flutpolder sind trockene oder teilweise vernässte Staubecken, welche typischerweise durch landwirtschaftliche oder andere Landnutzung geringer öko- nomischer und ökologischer Vulnerabilität gekennzeichnet sind. Ziel dieser Dissertation ist die Analyse von hydraulischen, ökologischen und ökonomischen Auswir- kungen des Einsatzes von Flutpoldern, um daraus Schlussfolgerungen für ihre Bewirtschaftung zu zie- hen. Dazu werden hydrodynamische und Wassergütemodelle eingesetzt sowie erfolgt eine Abschätzung der ökonomischen Vulnerabilität in zwei Untersuchungsgebieten an der Mittleren Elbe in Deutschland. Ein Untersuchungsgebiet ist ein existierendes Flutpoldersystem am Nebenfluss Havel, welches während der Elbeflut im Sommer 2002 zum Einsatz kam. Das zweite Untersuchungsgebiet ist ein geplanter Flut- polder, welcher sich bisher noch in einem frühen Planungsstadium befindet. Entsprechend können die Studien als ex-post und ex-ante Untersuchungen bezeichnet werden. Hydrodynamisch-numerische Modelle werden in beiden Untersuchungsgebieten angewandt, um die Hochwasserkappung und die Überflutungscharakteristik unter verschiedenen Hochwasserszenarien und Kontrollstrategien zu untersuchen. Es wird gezeigt, dass Hochwasserspitzen durch zeitweiligen Wasserrückhalt effektiv gekappt werden können. Jedoch belegen die Untersuchungen auch, dass der erzielbare Kappungsbetrag stark von der Form der Hochwasserwelle abhängt. Überflutungskenngrößen wie Wassertiefe, Fließgeschwindigkeit und Überflutungsdauer weisen eine große räumliche Variabilität innerhalb der Flutpolder auf. Diese Daten bilden eine Grundlage für die anschließende Wassergütemo- dellierung und Schadensabschätzung. Eine lang anhaltende Wasserspeicherung kann negative ökologische Effekte hervorrufen. Ein großes Problem, welches in Flutpoldern an der Elbe während des Hochwassers im August 2002 beobachtet wurde, stellt die Zehrung von gelöstem Sauerstoff aufgrund des hohen Sauerstoffbedarfs dar, der durch organisches Material im Wasser und auf landwirtschaftlichen Feldern hervorgerufen wird. Daher wird die Dynamik des gelösten Sauerstoffes während der Flutungs-, Stagnations- und Entleerungsphase des Flutpoldereinsatzes mit einem numerischen Modell simuliert. Die Abschätzung der ökonomischen Vulnerabilität konzentriert sich hauptsächlich auf landwirtschaftliche Flächen entsprechend der vorherrschenden Flächennutzung in den Flutpoldern. Im Gegensatz zu anderen Schadenkategorien weisen Verluste auf landwirtschaftlichen Feldern einen ausgeprägten saisonalen Verlauf auf, während sich die Hochwasserwahrscheinlichkeit auch im Jahresverlauf ändert. Beide saisonalen Aspekte werden bei der Abschätzung der jährlichen Verluste auf landwirtschaftlichen Flä- chen einbezogen. Darüber hinaus werden numerische Modelle verschiedener räumlicher Dimensionalität von null- bis zwei-dimensional angewandt, um ihre Eignung für hydrodynamische und Wassergütesimulationen von Flutpoldern hinsichtlich der Leistungsfähigkeit und des Modellierungsaufwands zu bewerten. Für viele Anwendungen liefern Modelle geringere Dimensionalität ausreichend genaue Ergebnisse, während sie die Untersuchung von vielfältigen Hochwasserszenarien und eine ausführliche Überprüfung der Mo- dellsensitivität ermöglichen. Im Gegensatz dazu benötigen zwei-dimensionale Modelle einen erheblich höheren Rechenaufwand, sind allerdings für bestimmte Anwendungen unumgänglich. Allgemein sollten zwei-dimensionale Modelle in Überflutungsgebieten mit komplexer Topographie und ausgeprägter räumlicher Variabilität von hydraulischen Größen wie Fließgeschwindigkeit und Wassertiefe bevorzugt 10 werden. Für die praktische Umsetzung wird vorgeschlagen, mit einem einfachen Ansatz zu beginnen und die räumliche Dimensionalität zu erhöhen, falls die Zielgrößen eine hohe räumliche Variabilität aufweisen. Die Dissertation schließt mit übergreifenden Empfehlungen zur Bewirtschaftung von Flutpoldern ein- schließlich Kontrollstrategien und Landnutzung ab. Im Hinblick auf zukünftige Änderungen in der Auftretenshäufigkeit von Hochwassern und weiterhin ansteigenden Werten von privatem und öffentli- chem Vermögen in überflutungsgefährdeten Gebieten stellen Flutpolder ein effektive und flexible Maßnahmen des technischen Hochwasserschutzes dar, welche zu einem erfolgreichen Hochwasserrisi- komanagement großer Tieflandflüsse beitragen.

Chapter I

Introduction ures such as hazard zoning, early warning systems or evacuations become most important. 1 BACKGROUND Flood polders are embanked areas that are intentionally inundated during large floods so as In the last few decades, floods in Europe to reduce peak flood flows and hence the haz- have become a growing issue of concern for ard in adjacent and downstream river reaches. citizens and authorities (Alphen et al., 2008) as Flood polders contribute to the flood risk man- several extreme river floods have occurred in agement strategy for many lowland rivers in many European countries including Germany. Germany, other European countries and else- At the same time, the value of public and pri- where (for examples, cf. Dijkman et al., 2003). vate assets in flood prone areas has further in- The term “flood polder” originates from the creased. Moreover, effects of climate change on equivalent German word. Other commonly flood frequency are projected in many studies. used terms are flood detention area or off- For instance, Christensen and Christensen stream flood storage reservoir. Figure 1 shows a (2003) state that the frequency of severe sum- schematic view of a flood polder. It is usually mer flooding over large parts of Europe is likely enclosed by a separation dike to the river and to increase, despite the general trend to drier by an additional polder dike to the hinterland if summer conditions. Furthermore, regional an increasing topography does not limit the wa- changes in the seasonality of floods have al- ter flow (Fischer et al., 2005). ready been observed in many areas of Europe Ideally, the discharge up to a pre-defined as a result of changes in the flood-generating level will be passed into the flood polder by use mechanisms (Kundzewicz, 2008). These devel- of adjustable control structures. For a maxi- opments have triggered a new comprehensive mum peak reduction, the flood wave needs to approach on flood risk management on a Euro- be cut horizontally. During the emptying proc- pean level which finally led to the adoption of ess, the stored water should be released as soon the Directive on the Assessment and Manage- as possible provided that water levels in the ment of Flood Risk (EU, 2007). According to main river allow for safe discharges at the this directive, flood risk is defined as a combi- downstream river reaches (Hall et al., 1993). nation of the probability of a flood event and of The land within the storage area may be used the potential adverse consequences to human for agriculture or other low-intensity purposes, health, the environment, cultural heritage and but is not suitable for buildings or similar in- economic activity associated with a flood event. vestments (Smith and Ward, 1998). Flood risk management comprises measures The effectiveness of a flood protection measure that aim to reduce flood hazard and vulnerabil- refers to the extent to which the measure’s ob- ity to floods. Certain flood risk measures apply jectives are achieved (OECD, 2002). Flood to floods of different recurrence intervals peak reduction, being the main objective of (DKKV, 2004). During rare floods with recur- flood polder utilisation, is influenced by several rence intervals between approximately 10 and factors such as the storage capacity compared 200 years, the flood hazard can be reduced by to the discharge volume of the flood wave, the technical measures such as dikes, reservoirs or operational scheme of the control structures, so-called flood polders, whereas during very the quality of the flood forecast and the shape rare floods with recurrence intervals larger than of the flood wave (Dijkman et al., 2003). approximately 200 years, organisational meas- 12 Introduction

Fig. 1: Schematic of flood peak attenuation by the use of controlled water storage in flood polders

The efficiency is a measure of how eco- impacts of the utilisation of flood polders for nomically resources are converted into results flood peak reduction during large flood events. (OECD, 2002). In the case of flood polders, it It is addressed by applying models of different is mostly affected by the prevented damage in complexities in terms of spatial dimensionality the downstream areas or reduced expenses for (hydraulic and water quality modelling) and flood management, the costs for construction time frame (vulnerability assessment) in order and maintenance of the flood polder dikes and to evaluate their suitability for flood polder control structures, damage occurring due to the studies. temporary water storage and the probability of To tackle the overall objective the following utilisation (Gocht and Bronstert, 2006). research questions are posed: Apart from economic losses, utilisation of  How effective are the flood polders in flood polders can lead to environmentally terms of peak reduction for varying harmful situations. An environmental side ef- flood scenarios and operational fect that has been observed during long storage schemes? times is the depletion of dissolved oxygen in  How does the long water storage affect the water due to the strong oxygen demand im- water quality and particularly dissolved posed by organic material in the water body and oxygen levels in the flood polders? on the agricultural fields. This may cause stress  How can economic vulnerability be as- on terrestrial and aquatic ecosystems and in par- sessed when considering time-varying ticular on fish populations in the storage area or damage in agricultural areas? in the river after release of stored water. After  Which is the appropriate model com- flood polder utilisation at a tributary of the plexity to simulate hydraulic and water Elbe River in summer 2002, an almost com- quality processes in flood polders? plete fish extinction in the adjacent river reach was observed, which was directly linked to the depletion of dissolved oxygen in the water 3 STUDY AREAS (Böhme et al., 2005). To address the research questions, investigations are carried out in two study areas, both 2 OBJECTIVE AND RESEARCH situated on the Middle Elbe River in Germany QUESTIONS (Figure 2). The first study area is a flood polder system on the tributary Havel (hereinafter called The overall objective of this thesis is to as- “Havelpolder system”), which was constructed sess hydraulic, environmental and economic in the 1950s and was set in operation during the Chapter I 13

Elbe flood in summer 2002. The Lower Havel is planned to be 40 million m³. The planned floodplain and six flood polder reservoirs on flood polder consists of two separate reservoirs. both sides of the Havel River near its conflu- Adjustable gates will control the inflow from ence with the Elbe River provide room for re- and outflow to the river and the flow between taining the Elbe flood peak. The maximum the two reservoirs. storage capacity of the flood polder reservoirs Both flood polder systems are designed to amounts to 110 million m³, while the overall reduce flood peaks having a return period of storage capacity of both the flood polder reser- approximately 100 years or larger, i.e. they will voirs and the floodplain is approximately 250 only be operated during very large flood events. million m³. Adjustable inlet control structures The two study areas have a similar land use, are only installed in one of the reservoirs. In- which is dominated by intensive arable land and stead, the other reservoirs were opened by dike grassland. The agricultural fields constitute a blasting or excavation during their utilisation in large source of degradable organic material summer 2002. when inundated. After designation as flood The second study area is the planned flood polder, the original land use is maintained and polder system Axien, which is not yet existing farmers are paid for their losses in the case of but in the early planning stages by the federal flood polder utilisation. water authorities. Its maximum storage capacity

Fig. 2: Map of the Elbe catchment with locations and photographic impressions of the study areas (modified after IKSE, 2005)

The Elbe is the forth largest river in Central The runoff seasonality of the Elbe River is Europe, having a total length of 1092 km and a strongly influenced by the Czech and German catchment area of nearly 150000 km², which is middle-mountain regions which cover about shared by Germany (65 %), the Czech Republic 30 % of the catchment area. It can be character- (34 %), Poland and Austria (together less than ised as a pluvio-nival regime with highest dis- 1 %). The main tributaries are the charges in March and April and low discharges Moldau/Vltava, Saale and Havel, each compris- from June to November. Compared to the run- ing a catchment area of approximately off behaviour of some other large European 25000 km² (Figure 2). rivers, e.g. the River Rhine, discharge in the 14 Introduction

Elbe River is not affected by stored water in 4 OUTLINE OF THE THESIS glaciers and permanent snow. More than 80 % of the floods occur during the winter and Figure 3 illustrates how the chapters are re- spring season in consequence of snow melt as- lated to the main research subjects, i.e. the in- sociated with intense rainfall. Floods in winter vestigation of hydraulic, environmental and or spring are characterised by long durations, economic impacts of flood polder utilisation. while summer floods caused by heavy continu- Chapter II deals with the hydraulic simula- ous rains typically only last a few days, but they tions and the economic analysis carried out in usually have higher flood magnitudes (IKSE, study area 1. Effects of different flood scenar- 2005). ios and operational schemes on flood polder The present-day embankments confining utilisation in study area 2 are investigated in most of the German part of the Elbe River date Chapter III, while Chapter IV focuses on the back to the second half of the 19th century. comparison of hydrodynamic models of differ- However, diking at the Elbe River started as ent spatial dimensionality for the same study early as the 12th century. The dike construction area. Chapter V investigates the impact of flood led to a reduction of the retention area on polder utilisation on dissolved oxygen dynamics German territory from 6172 km² to 838 km² in study area 2, whereas economic vulnerability (13.6 % of the original area) (BfG, 2002). To- is estimated in Chapter VI. And finally, chapter day, the reactivation of floodplains by dike- VII presents overall conclusions based on the shifting and the designation of flood polders several studied aspects and gives recommenda- are discussed. Several sites along the middle tions regarding flood polder management and course of the Elbe River have already been future research. suggested as potential locations for flood pol- Chapters II to VI were written as stand-alone ders (IKSE, 2003) and were investigated in manuscripts that are either published or await- terms of flood peak reduction potential (Helms ing publication in international peer-reviewed et al., 2002; Busch and Hammer, 2007). journals (for full reference, see front pages of the chapters).

Chapter I: Research background, objective and overview

Study area 1 Study area 2 Havelpolder system Flood polder Axien

Model comparison

Chapter II: Chapter III: Chapter IV: Hydraulic Hydraulic simulations Comparison of Hydraulic simulations for for different flood 1D and 1D-2D different flood scenarios and hydraulic models scenarios operation schemes

Chapter VI: Cost-benefit Assessment of Economic analysis for economic different flood vulnerability in scenarios rural areas

Chapter V: Comparison of Environ- 0D and 2D water mental quality models

Chapter VII: Discussion, conclusions and recommendations

Fig. 3: Outline of the thesis

Chapter II

Flood risk reduction by the use of retention areas at the Elbe River

ABSTRACT: The paper presents research results on flood risk mitigation by the controlled flooding of a retention area on the middle reaches of the Elbe River. The retention area consists of six large polders and the floodplain of a tributary, the Havel, and is located near the Havel’s confluence with the Elbe River. The total retention volume of both the polders and the Havel floodplain amounts to approximately 250 million m³. The controlled flooding of the retention area was simulated by the use of a conceptual model and assessed economically for two flood scenarios. In a cost-benefit analysis, the damage to agriculture, the road network, buildings and fishery caused by the flooding of the polders was compared with the resulting reduction in potential damage in the town of Wittenberge, 30 km downstream. On the basis of a monetary assessment it was concluded that the use of the retention area for flood protection is highly cost-effective in economic terms.

Published as: Förster S, Kneis D, Gocht M and Bronstert A. 2005. Flood Risk Reduction by the Use of Retention Ar- eas at the Elbe River. Intl. Journal of River Basin Management, 3 (1), 21-29.

16 Flood risk reduction by the use of retention areas at the Elbe River

volume of the Havel floodplain and polders together amounts to approximately 250 million m³. 1 INTRODUCTION

In the last few years, floods have caused enor- mous damage in Central Europe. The flood of the Elbe River and its tributaries during the summer of 2002 was accompanied by the highest water levels ever measured at many gauging sta- tions. The overall damage in Germany amounted to about 10 billion € (DKKV, 2004). Flood risk management aims at minimising the impact of flood disasters (Plate, 2002). The use of retention areas can be an efficient measure in modern flood risk management. By controlled flooding of sparsely or non-populated areas with relatively low damage potential, the risk of inundation for downstream areas with higher vulnerability can be reduced. Fig. 1: Simplified map of the retention area near the con- The largest retention area along the Elbe River fluence of the Havel and the Elbe River is situated on the tributary Havel near its confluence with the Elbe. This retention area consists Before the extensive construction of dikes and of six large polders and the floodplain of the water-engineering works, the whole Lower Havel Lower Havel River which together have a poten- River floodplain was a natural retention area for tial retention volume of up to 250 million m³. the Elbe River and was characterised by frequent The system was constructed in the 1950s, but inundations. The dikes were constructed in order was used operationally for the first time during to protect settlements and agricultural areas from the Elbe flood in 2002. By controlled flooding of flooding. Extensive melioration and deforestation the retention area the peak stages were attenuated in the last centuries have enabled an intensive ag- by approximately 40 cm at the gauge of Witten- ricultural use of the polder area. The farmland is berge (BfG, 2002). Consequently, the risk of in- used as arable land and grassland in approxi- undation for the town of Wittenberge and areas mately equal shares. The main crops are winter further downstream was significantly reduced. grain and corn. The polders are only sparsely The flood peak reduction resulting from the populated. use of the retention area was considered very suc- The discharge of the Havel into the Elbe cessful by the water authorities. However, this as- River is controlled by several gates. In the case of sessment was not based on an economic evalua- an extreme flood along the Elbe River, the gates tion including costs and benefits of the measure. are closed in order to prevent the uncontrolled Therefore, the aim of this study was to assess inflow of water from the Elbe into the Havel economically the use of the polders described River. However, in time of peak discharge on the above for flood protection in order to gain valu- Elbe River, an emergency gate can be opened in able information for their further use. order to divert and temporarily retain the Elbe flood peak in the retention area. The resulting flood peak reduction mitigates the flood risk for 2 DESCRIPTION OF THE STUDY AREA areas further downstream. Within this study the benefit of flooding the retention area for the In the case of an extreme flood event along town of Wittenberge was investigated. This town the middle stretches of the Elbe River, both the is situated on the Elbe about 30 km downstream floodplain and the polders at either side of the of the confluence, and has a population of ap- Havel River provide storage capacity for retaining proximately 25 000 inhabitants. The lower parts the Elbe flood peak (see Figure 1). The six pol- of the town are protected against floods by em- ders comprise an area of 100 km² with a volume bankments and mobile walls (see Figure 7). of 110 million m³. The overall potential retention Chapter II 17

3 RESEARCH APPROACH flood peak was not attenuated. Figure 2 shows the overall investigation scheme of the study. Two different flood scenarios were analysed in this study. As a reference scenario the Elbe flood of August 2002 with a maximum discharge of 4225 m³/s at the gauge at Wittenberge was selected, since detailed hydrological observation data were available. Using these data, a conceptual inundation model (section 4) was calibrated and tested. The model provides a means to simulate the flooding of the retention area for the 2002 event but it also allows further user-defined flood scenarios to be investigated. The return period of the reference scenario was estimated to be about 180 years based on the discharge series 1900–2002. A more extreme scenario (referred to as “scenario II”) was derived from the 2002 flood event by increasing the ordinates of the observed dis- Fig. 2: Investigation scheme used within this study charge hydrograph by a factor of 1.05. This trans- formation not only raises the peak discharge from 4225 to 4440 m³/s but also increases the 4 MODELLING THE INUNDATION OF total volume of the flood wave. The intention of THE POLDER SYSTEM scenario II was to design the largest possible flood that would not cause embankment failures The estimation of damage in the polders pri- in the town of Wittenberge provided that the marily requires information on the inundated area polder system is used for flood peak retention. and the duration of flooding. While for the 2002 The return period of the second flood scenario reference scenario these data could be derived was estimated to be about 300 years. from gauge observations and aerial views, the in- In the first step of the economic assessment it vestigation of scenario II requires a simulation of was assumed that the polder system is used for the inundation process. For that purpose a simple flood peak retention and consequently embank- but robust conceptual model, the “Coupled Res- ment failures in the town Wittenberge do not oc- ervoirs Model” (short: CR Model), was devel- cur. For both flood scenarios stage hydrographs oped. Section 4.1 introduces the general outline for the Havel River and the adjacent polders were of this model whereas section 4.2 deals with its deduced either from observed data (for the flood application to the retention area on the Lower in 2002) or model simulations (for scenario II). Havel River. The inundation extent and duration of flooding were determined from these hydrographs. Com- 4.1 Outline of the CR Model bining the results with data on agricultural land use and on the asset of houses and roads enabled In the CR Model, the natural system consist- the calculation of damage within the polders for ing of river sections, floodplains and polders is both scenarios (section 5). discretised into a number of reservoirs. It is as- In a subsequent step it was hypothesised that sumed that the water surface slope within each the polders are not flooded, resulting in a partial reservoir is negligible. Hence, reservoirs are de- inundation of Wittenberge. Therefore, potential scribed by simple storage functions relating water damage in this town was determined for the two surface elevation to storage volume, mean depth flood scenarios (section 6). and inundated area. Each reservoir may be linked Finally, a cost-benefit analysis was carried out to one or more adjacent reservoirs, allowing wa- (section 7). Within this analysis losses resulting ter to be exchanged. Boundary conditions can be from controlled flooding of the polder system assigned to reservoirs in order to account for ex- were compared with the potential damage in the ternal inflows and outflows of the modelled sys- town of Wittenberge that would occur if the Elbe tem. Figure 3 exemplifies the model outline showing two polders connected to a river. The river 18 Flood risk reduction by the use of retention areas at the Elbe River reach with its associated floodplain is subdivided into sections, each represented by a discrete but   BO GW  hhif)hh( B interconnected reservoir. The system’s in- and Qs LA   outflows are associated with the most upstream   GWO GW  hhif)hh( B (1) and downstream river section, respectively. Two auxiliary reservoirs allow for the storage of water in the polders’ soil zones. -1 where QS = seepage flow (m³ s ), A = inundated area of the surface water reservoir (m²), hO = surface water elevation (m), hGW = groundwater surface elevation (m), hB = average terrain elevation of the inundated area (m). The leakage factor L (s-1) is subject to calibration. For reasons of simplicity it is assumed that seepage causes the ground water surface to rise instantaneously with the soil water content of the above layer remaining unchanged. Furthermore the groundwater surface is assumed to be horizontal and lateral groundwater flow between reservoirs is not simu-

Fig. 3: Scheme of a simple polder system illustrating the lated. outline of the Coupled Reservoirs Model. See text for ex- Two types of boundary conditions are avail- planation of individual elements. able in the CR Model for representing inflows and outflows of the modelled system. On the one The flow between coupled surface water res- hand external in-/outflows can be assigned to a ervoirs may be calculated by two different ap- reservoir by directly specifying a discharge time proaches. The first option is to compute flow series. On the other hand stage hydrographs can rates from both slope and water surface elevation be used as boundary condition. In the latter case at the interface cross section of adjacent reser- the reservoir’s in-/outflow rates are derived from voirs. The underlying assumption is that a con- a pre-processed lookup table, relating the flow to tinuous slope exists between the coupled reser- both the external stage given by a time series and voirs, though in the model each reservoir is the stage in the modelled reservoir itself. represented by a single stage only. Hence, the wa- The output files of the CR Model provide ter surface elevation at the interface cross section time series of water level and storage volume for can be estimated from the reservoirs’ stages. This each reservoir. Furthermore, flow rates between option provides a simple means to compute the reservoirs (link flows) and boundary conditions flow between river sections, each of them repre- are continuously recorded. sented by a separate reservoir (see “Type 1 link” in Figure 3). During the simulation, pre- processed lookup tables are used to determine 4.2 Application of the CR Model to the Lower the current flow rate depending on gradient and Havel River stage. A second option is used if adjacent reser- The retention area at the Lower Havel River voirs are connected by hydraulic control struc- was modelled as a system of 17 reservoirs (see tures. Because the water surface profile changes Figure 4). The Havel River with its associated rapidly at these locations, flows are directly calcu- floodplain was discretised into six subsections H1 lated from the reservoirs’ stages. Control struc- to H6. The remaining 11 reservoirs include the tures like breaches, weirs and culverts usually six large emergency polders, two of them subdi- provide the link between polders and the river vided into two separate basins (P1–P8). The la- (see “Type 2 link” in Figure 3). Again, lookup ta- bels P9 to P11 mark three smaller polders sepa- bles are used to derive the flow from stage in- rated from the floodplain by small dams only. formation during a model run. Storage functions for all reservoirs were deduced A special type of reservoir was implemented from digital elevation models in a pre-processing to account for the storage of water in the unsatu- step with the results being stored as lookup tables rated soil zone (see Figure 3). Seepage from sur- for subsequent use by the CR Model. Auxiliary face water reservoirs to the aquifer is estimated reservoirs (not shown in Figure 4) represent the using a leakage approach (Eq. 1): unsaturated soil zone of polders and floodplains Chapter II 19 sections. Relationships between ground water mented by assigning inflow hydrographs to the level and storage volume for these reservoirs corresponding reservoirs, a stage boundary con- were estimated from the soil’s air-filled porosity dition was used to simulate the system’s outflow at field capacity and terrain elevation data. through a constructed channel (label B4 in Figure 4). This was necessary as the outflow depends on both the stage at the channel’s confluence with the Elbe River and the simulated stage at the channel’s upstream end (reservoir H1). In addition, evaporation losses were taken into account in the simulation of the reference scenario to re- flect the evaporation caused by the exceptionally hot weather conditions in August 2002. As depicted in section 4.1, the CR Model ac- cesses pre-processed lookup tables to compute the flow between adjacent reservoirs. The crea- tion of appropriate tables was a major challenge in setting up the model. Connections between the floodplain sections H1 to H6 (Figure 4) were represented by links of “Type 1” (see section 4.1). Hence, relations between flow, water surface gradient and stage at the interface of the reservoirs had to be identified. For the reservoirs H1 and H2 a regression model was fitted, since stage and discharge are continuously recorded at the Fig. 4: Coupled Reservoirs Model of the retention area on Havelberg gauge (Figure 4) and stage information the Lower Havel River. Reservoirs that correspond to polders and floodplain sections of the Havel River are labelled is also available from adjacent gauges. The re- P1–P11 and H1–H6, respectively. Dikes separate the pol- gression is of the type: ders from the floodplain. Boundary conditions are labelled B1 to B4. The actual confluence of the rivers Elbe and Havel (not shown on the map) is located 8 km northwest  S)h(PQ (2) of symbol B4. Further details are given in the text.

All boundary conditions are shown in Fig- where Q = flow (m³ s-1), S = slope of the wa- ure 4. The inflow of the Havel River into the ter surface (–) estimated from the stage differ- modelled system is labelled B1. Discharge is con- ence of adjacent gauges, and P(h) is a polynomial stantly measured by an ultrasonic flow meter at with the stage h (m) as argument. The goodness this location providing reliable input data for the of fit is shown in Figure 5. simulation of the 2002 reference scenario. For simplicity, a constant inflow at B1 was assumed for the simulation of scenario II. All minor tribu- 200 Q comp = Q obs

taries of the Havel River are labelled B2. Due to [m³/s] 100

the lack of flow gauges, their discharges had to be comp estimated from catchment size using mean spe- 0 cific discharges of nearby river basins. B3 marks -100 the emergency gate through which water from -200 the Elbe can be diverted into the Havel River. During the 2002 flood event, water levels were -300 recorded by the operators and the flow through -400

the gate was computed using a weir equation. For Q discharge Computed -500 scenario II a corresponding flow series was gen- 0 100 200 erated which reflected the desired discharge re- -500 -400 -300 -200 -100 duction in the Elbe River as well as basic hydrau- Observed discharge Q obs [m³/s] lic features of the inlet structure and the overall Fig. 5: Observed and computed discharge at the Havelberg storage capacity of the retention area. Whereas gauge (see Figure 4) using multiple regression the boundary conditions B1 to B3 were imple- 20 Flood risk reduction by the use of retention areas at the Elbe River

Figure 5 indicates that the flow rate may be map for a single polder. The map was generated predicted from the regression with an acceptable from a simulated stage hydrograph in a two step level of error. This is true for both normal condi- procedure. Firstly, grid maps of the inundated tions (Q > 0) and reverse flow. The observed area were created for each day by reclassifying the negative discharges at Havelberg occurred when polder’s digital elevation model. A value of 1 was the emergency gate (label B3 in Figure 4) was assigned to inundated grid cells and the remain- opened for flood peak attenuation in 2002. ing dry cells were set to zero. Secondly, the grid Lookup tables which are required to predict the maps of all successive days were summed up to flow between the remaining floodplain sections yield the spatial pattern of flood duration. Similar H2 to H6 (Figure 4) were derived from cross sec- maps as shown in Figure 6 were produced for all tion geometry data using Manning’s law. Rough- polders P1–P8. ness coefficients were taken from previous applications of a 1D hydrodynamic model. However, the computed flow rates should be considered as rough estimates only. Since the polders P2–P11 (Figure 4) are flooded either through dike breaches or weirs, their connection to the corresponding floodplain sections was implemented using “Type 2” links (see section 4.1). The flow through either inlet structure was calculated using the POLENI equation:

2 2   hwgcQ 2/3 (3) 3 u Fig. 6: Flood duration map of the polder Twerl (label P6 in Figure 4) for scenario II as it was derived from a stage hydrograph simulated by the CR Model. The dike breach where Q = flow (m³ s-1), µ = overflow coeffi- through which the polder was flooded is located in the cient (–), c = submergence correction factor (–), middle of its southern boundary. w = width of the inlet (m) and hu = upstream energy head (m) above the crest. Appropriate values 5 DAMAGE ASSESSMENT IN THE of µ were taken from (FAS, 1975) or calibrated in POLDER SYSTEM case of the breaches. The submergence correction factor c was estimated according to Schmidt Damage to agriculture, the road network, (1957), using the ratio of downstream to up- housing and fishery was assessed on a monetary stream water level above the crest. Approaches of basis as described in the following. pipe hydraulics were used to create the required Agricultural damage strongly depends on the lookup table for polder P1 (Figure 4) since cul- time of occurrence of the flood event within a verts provide the link to the Havel River at this year (see Table 1). In general, floods occurring location. before the sowing of the spring grain cause least Simulations were carried out with a computa- damage whereas maximum losses are caused by tional time step of 15 min. Since flow rates inte- flood events occurring shortly before harvest. grated over the length of a time step are small Also, losses vary greatly between the agricultural compared to the reservoirs’ storage volumes, the land use types. Therefore, the damage to agricul- CR Model uses a simple explicit solution tech- ture was calculated for both flood scenarios de- nique. pending on the extent and duration of inundation, the spatial distribution of the land use types 4.3 Aggregation of simulation results and the timing of flood occurrence. Damage to roads depends on the extent of the Since damage assessment in the polders (sec- inundated area, the spatial distribution of road tion 5) is based on inundation maps only, hydro- construction types and the degree of damage. graphs and further results of the scenario simula- Unfortunately, cost assessments for the 2002 tions using the CR Model are not presented here. event were not made available by the communi- Figure 6 gives an example of a flood duration ties affected by the flood. Therefore, costs for Chapter II 21 repair or reconstruction of the total road network ries appeared inappropriate given the low ex- within the polder area had to be calculated on the pected amount of loss. Thus, further categories basis of bid prices (see Table 2). In the cost- were not included in this study. benefit analysis (section 7) calculated repair and reconstruction costs were used as lower and upper bounds of the total damage to the road net- 6 ASSESSMENT OF POTENTIAL work, respectively. DAMAGE REDUCTION IN WITTENBERGE Tab. 1: Calculated agricultural losses in the polders for the two flood scenarios (million €) The benefits of using the polders appear downstream in the form of prevented damage. Flood Scenario I Scenario II There is a large range of damage types to be con- occurrence sidered. Generally, direct damage resulting from Before sowing of 1.47 2.26 the physical contact of the floodwaters with hu- spring grain (mid mans or properties and indirect damage like dis- April) ruption of traffic and trade can be distinguished (Smith and Ward, 1998). Further differentiation After sowing of 2.40 3.70 into tangible and intangible damage is possible. spring grain (mid Besides the prevention of flood damage, changes April) in defence and evacuation effort as well as dike Before harvest 3.08 4.73 restoration costs can be taken into account. The scope of this investigation was rather the (mid July) development of a comprehensive approach to- After harvest (mid 2.20 3.39 wards assessing the use of polders for flood re- July) tention than a detailed study of all damage and benefit categories. Therefore, only prevented di-

Tab. 2: Calculated damage to roads in the polders for the rect damage to industry, trade and services, pub- two flood scenarios (million €) lic facilities and housing in the downstream town of Wittenberge was considered, which is referred Type of costs Scenario I Scenario II to as “potential damage” in the following. The first step in the assessment was to deter- Repair costs 3.26 3.59 mine the extent of inundated areas for both flood Reconstruction 5.16 5.66 scenarios (section 6.1). Subsequently, the corre- costs sponding damage was calculated as described in 6.2.

Due to the small number of houses in the retention area, damage to buildings was estimated 6.1 Determination of flooded areas to be low compared to the other damage catego- The town of Wittenberge is protected against ries. On the basis of observation and experience, floods by embankments. Over a total length of estimates range from 0.17 to 0.22 million € for 1100 m in the harbour area of Wittenberge, these scenario I and from 0.32 to 0.39 million € for embankments do not reach the full height re- scenario II. Losses to fishery were ascertained by quired by the design flood for reasons of amen- the appropriate authorities for compensation pur- ity. As a makeshift, mobile walls are used for poses. The costs amounted to 0.47 million € for flood protection at this location. These walls are the reference scenario. estimated to have a larger failure probability than The damage categories described above were the embankments. considered to be most relevant in terms of a monetary damage assessment. The effort involved in assessing the damage to further catego-

22 Flood risk reduction by the use of retention areas at the Elbe River

Fig. 7: Flood hazard map for the town of Wittenberge (flood scenario II with an estimated return period of about 300 years)

The use of flood polders reduces the failure The volume of the embankment overflow at probability of flood protection measures down- the failure site was calculated using the weir for- stream. In a simplified scenario it was assumed mula of POLENI (Eq. 3). For the flood scenar- that the mobile flood protection walls would fail ios defined in section 3, volumes of 7.4 million over a length of 500 m in the harbour of Witten- m³ for the reference scenario and 21.6 million m³ berge and that overtopping of the embankments for scenario II were estimated. would occur unless the polders were used for In order to determine the affected area in the flood peak retention. The volume of the flood town of Wittenberge a relationship between the wave exceeding the embankment’s crest level was water level and the corresponding storage volume assumed to flow through the lower parts of Wit- was established from the digital elevation model. tenberge. The failure site and the flooded area for Using this relation the inundated area and the wa- scenario II are shown in Figure 7. Because occur- ter level behind the failure site was derived for rence probability and intensity of flooding are the estimated outflow volumes of 7.4 and 21.6 known for the scenario given in the map, it is re- million m³ (Figure 8). ferred to as hazard map (Merz et al., 2005). 35

30 Reference Scenario 25 Scenario II

5 volume in Wittenberge [ million m³]

0 19.0 19.5 20.0 20.5 21.0 21.5 22.0 22.5 water level [m above msl] Fig. 8: Inundation volume in the town of Wittenberge against inundation water level behind the failure site as result of hydraulic calculations Chapter II 23

each land use type. The potential damage per land use unit depends on the water depth. Rela- 6.2 Assessment of potential damage in tive stage-damage functions, giving flood damage Wittenberge as a percentage of an area’s total value against wa- Direct damage to industry, trade and services, ter depth, were used to determine damage in cur- public facilities and housing was estimated on the rency units. The applied method was based on a basis of land use. The German land register AT- reference study from the German federal state of KIS served as the spatial basis for land use in- North Rhine-Westphalia (MULR, 2000). A more formation. An analysis of the national accounts detailed description of the approach is included delivered specific economic values (€/m²) for in the project report (Bronstert, 2004).

Fig. 9: Flood risk map for the town of Wittenberge (flood scenario II). Potential damage is given for each flood affected land use unit.

The intersection of land register and water tial flood damage of about 17 million € was esti- depth enabled the calculation of the potential mated. Scenario II resulted in potential damage damage for the areas affected by flooding. of about 31 million €. Figure 9 shows a risk map where potential damage for each land use unit is mapped. Be- cause hazard and vulnerability are known for the 7 COST-BENEFIT ANALYSIS scenario shown in the map, it is referred to as risk map (Merz et al., 2005). The cost-benefit analysis aggregates those pro- Uncertainty of damage assessment was con- ject results which are expressed in monetary units sidered by varying the specific economic values into a cost-benefit ratio. Because a complete for each land use type on the basis of an on-site monetary account of all benefit and cost posi- inspection of the area, yielding a lower and upper tions could not be elaborated within the project, boundary of the potential damage. the ratio cannot serve as the only decision crite- The result of the damage assessment in the rion. However, a comprehensive assessment form of a graphical representation of potential should be based on a decision support taking all flood damage against inundation volume is given relevant perspectives into account. in Figure 10. For the reference scenario, a poten- 24 Flood risk reduction by the use of retention areas at the Elbe River

Fig. 10: Potential damage against inundation volume in the town of Wittenberge

The polders on the Lower Havel River were The large spread of the benefit-cost ratio is a built as a multi-purpose system. Besides flood result of combining low benefit and high cost es- protection of downstream riparian areas they en- timates and vice versa. The use of lower and up- able agricultural land use and protect infrastruc- per estimates of cost (section 5) and benefit (6.2) tural facilities. For the performance of a cost- in the calculation of the benefit-cost ratio serves benefit analysis a clear decision as to which costs as a pragmatic means to provide an idea of the are to be associated with a certain purpose is re- associated uncertainty. quired. Therefore, costs are differentiated into The high benefit-cost ratios indicate that for event-dependent ones and event-independent the investigated scenarios the use of the polders ones where “event” refers to “flood protection of for flood protection is economically highly bene- downstream riparian areas”. ficial. Because both benefits and costs are event- Event-dependent costs are damage to agricul- dependent, a variation in the anticipated lifetime ture, road network, buildings and fishery. Fur- of the system or discount rate would not affect thermore, the costs of blowing up and recon- the result. structing the dikes fall into this category as most It can be argued that each of the above- of the polders are not equipped with inlets. mentioned purposes should carry a certain por- Event-independent costs arise from constructing tion of construction and maintenance costs of and maintaining the dikes. Only event-dependent the polder dikes. A partial inclusion of these costs were included in the efficiency assessment event-independent costs would lower the benefit- of using the polders for flood protection. cost ratio. Furthermore, the result would become The lifetime of the system was assumed to be sensitive to changes in anticipated lifetime or dis- 90 years. A discount rate of 3 % was selected for count rate. taking time preference into account. As Table 3 indicates, the benefit-cost ratio calculated for scenario II is significantly higher than for the ref- 8 DISCUSSION AND CONCLUSIONS erence scenario. The aim of the study was to assess economi- Tab. 3: Calculated benefit-cost ratios of using the polder cally the use of a polder system for flood protec- system for attenuation of the Elbe flood peak tion. For this purpose an interdisciplinary research approach was essential since aspects of Benefit-cost Scenario I Scenario II hydraulics, agricultural science and economics ratio had to be accounted for. Lower estimate 1.5 2.2 From our investigation, it can be concluded that for the two extreme flood scenarios under Upper estimate 4.0 5.8 consideration, the controlled flooding of the re- Mean 2.8 4.0 tention area on the Lower Havel River is highly

cost-effective. For scenario II a higher benefit- cost ratio was calculated than for the reference Chapter II 25 scenario. This is mainly due to the fact that the polders as a flood protection measure in a holistic total potential damage in the town of Witten- way. In the analysis of costs and benefits (section berge varies greatly between the two flood sce- 7) it was assumed that the mobile wall in the har- narios. The higher inundation depth in scenario bour of Wittenberge would fail unless the polders II results in greater damage compared to the ref- were used for flood peak retention. However, a erence scenario, since loss is strongly correlated more comprehensive approach would entail a to inundation depth for most urban land use probabilistic failure assessment based on the reli- categories. However, losses in the polders show ability of flood defences downstream of the con- less variation between the two flood scenarios. fluence of Elbe and Havel. The larger volume retained in the polders in sce- Future research work should account for fur- nario II results in a higher inundation depth ther damage categories than were included in the rather than in an enlargement of the flood- current cost-benefit analysis. In this study only affected area. Since damage to agriculture and the direct tangible damage was considered. However, road network depends on the extent of the for a comprehensive evaluation of flood man- flooded area rather than on the inundation depth, agement strategies indirect as well as intangible the difference in total damage for the two flood damage should be taken into account. Whereas scenarios is comparatively low. approaches exist for the inclusion of indirect Experiences from the 2002 flood event as well damage like disruption to traffic, production or as from the conducted study proved that polders trade, an appropriate assessment of intangible of large storage capacity are essential to effec- damage such as damage to human health and tively attenuate an extreme flood peak of the ecological side effects is a major challenge for Elbe River. However, the full benefit of the stor- further research. age capacity can only be achieved by controlled operation and good timing. Modelling the inundation dynamics in the ACKNOWLEDGEMENTS polder area for the two considered flood scenarios was an essential part of the research. The CR We thank Werner Sauer (state office for envi- Model, which was developed for this purpose ronment of the federal state of Brandenburg), (section 4), proved to be an adequate simulation Gert Neubert and Ronald Thiel (state office for tool adapted to the system’s complexity and the agriculture of the federal state of Brandenburg), data availability. Apart from the two scenarios fo- Holger Ellmann and Burkhard Schulze (Ell- cused on in this study, the model is applicable to mann&Schulze GbR) for their contribution to various other flood scenarios. Thus, the effects the joint research project, which was funded by of initial and boundary conditions (e.g. varying the German Federal Ministry of Education and discharge of the Havel River) and of model con- Research. figuration (e.g. exclusion of selected polders) on Comments by two anonymous referees on an the inundation dynamics could be investigated. earlier version of this paper are gratefully ac- Throughout this study it emerged that further knowledged. research is required towards assessing the use of

Chapter III

Hydrodynamic simulation of the operational management of a proposed flood emergency storage area at the Middle Elbe River

ABSTRACT: Emergency storage areas can be an effective structural flood protection measure. By their controlled flooding the risk of inundation for downstream areas with higher vulnerability can be reduced. In the present study, the flooding and emptying process of a proposed storage area at the Middle Elbe River is simulated. The storage area has a maximum capacity of 40 million m³ and is divided into two polder basins. It is designed for the attenuation of extreme floods of 100 years or more return period. A one-dimensional hydrodynamic model is set up for a 20 km reach of the Elbe River, wherein the storage area is schematised by two storage cells each representing one polder basin. Flow between the storage cells and the Elbe River is controlled by adjustable gates which operate based on pre-defined conditions. Four flood scenarios which differ in flood magnitude and hydrograph shape are simulated. The scenarios are derived from analyses of a 70 years discharge record. Furthermore, for each flood scenario two gate control strategies are investigated. The results show that during large floods the utilisation of the storage area with controlled gate operations significantly reduces the Elbe River peak discharges. However, the magnitude of the attenuation depends on the steepness of the flood hydrograph and the applied control strategy with well-timed gate operations.

Published as: Förster S, Chatterjee C and Bronstert A. 2008. Hydrodynamic Simulation of the Operational Manage- ment of a Proposed Flood Emergency Storage Area at the Middle Elbe River. River Research and Applications. Pub- lished online, DOI: 10.1002/rra.1090.

28 Hydrodynamic simulation of the operational management …

potential detention areas along the Elbe have been proposed by the International Commission 1 INTRODUCTION of the Protection of the Elbe (2003) and their general suitability in terms of peak attenuation Flood emergency storage areas serve the pri- has been investigated (Helms et al., 2002, Büchele mary purpose of temporary water storage during et al., 2004). large flood events in order to reduce the flood One of these proposed detention sites has peak levels and thus alleviate the risk of inunda- been chosen for a detailed investigation in the tion in downstream areas with higher vulnerabil- present study as it seemed especially suitable for ity. Such storage areas are typically located along flood water storage because of its topography the middle reaches of large rivers. As a structural and location. It is situated in a river reach where flood protection measure they can form part of a several major dike failures occurred during the modern flood risk management system. flood in August 2002 resulting in a flooded area Storage is most effective if it is controlled, of more than 200 km². Due to the dike failures while the time of opening of the control gates is the flood peak was reduced by 11 cm (corre- crucial for a successful operation. This aspect has sponding to 220 m³/s) at the town of Witten- been pointed out by several authors (Jaffe and berg, situated about 30 km downstream of the Sanders, 2001, Aureli et al., 2005, Galbáts, 2006, proposed detention site (IKSE, 2004). This dem- Rátky and Szlávik, 2001). If the storage is utilised onstrates the potential of controlled water storage too early, most of the detained volume is taken at this location in reducing the risk of dike fail- from the rising limb. If the inlet gate is opened ures as well as lowering the water level in inun- too late, the volume is merely taken from the fal- dated areas just downstream along the river, ling limb. In both cases the obtained peak lower- where sites of cultural and industrial importance ing will be less than potentially possible (Silva et are situated. al., 2004). Ideally the flood hydrograph should be The objective of the present study is therefore cut to a constant discharge level while using the to determine the maximum peak attenuation full storage capacity (Figure 1). After the flood which can be obtained by the investigated poten- peak has passed by, the storage area should be tial storage area and to study the effect of differ- cleared of stored water as soon as water levels in ent floods and operation schemes on the peak at- the main river allow for a safe release (Hall et al., tenuation. This is done by detailed simulations of 1993). Besides a well-timed gate operation, the the filling and emptying process using the hydro- amount of peak attenuation depends on the char- dynamic model MIKE 11. Based on an analysis acteristics of the flood wave. of the 70 years discharge record at the Torgau gauging station different flood scenarios with regard to flood magnitude and hydrograph shape are derived for further investigation. Further- more, for each flood scenario two gate control strategies are considered. Finally, general aspects of controlled detention in emergency storage areas are discussed.

2 STUDY AREA

Fig. 1: Effect of ideal peak capping by controlled detention The proposed emergency storage area is lo- on hydrograph (dashed line). The unaffected hydrograph is cated in the lowland area at the Middle Elbe shown by a solid line. River in Germany between the gauges Torgau and Wittenberg. Figures 2 to 4 show an overview During the large Elbe flood in summer 2002 map of the study area, a cross sectional view of the effectiveness of emergency storage was evi- the modelled river stretch and of the polder ba- dent. By temporary water detention in the large sin, respectively. The storage area extends 7 km storage area at the confluence of Havel and Elbe along the right bank of the Elbe River. The over- Rivers, the Elbe flood peak was lowered by ap- all area amounts to 17 km² with a maximum stor- proximately 40 cm (Förster et al., 2005). Loca- age capacity of approximately 40 million m³. tions for further controlled and non-controlled Chapter III 29

The storage area is divided into a northern area is controlled by two adjustable overflow and a southern polder basin by an already exist- weirs of 25 m width each. Both weirs are divided ing dike road running through the area. Both ba- into four parts of 6.25 m width, which can be sins are connected by a sluice gate of 50 m width. operated independently. The filling and emptying process of the storage

Fig. 2: Map of the proposed flood emergency storage area. The Elbe-km information refers to the river mileage in German territory (starting at Elbe-km 0.0 at the Czech-German border)

Discharge characteristics of the gauges Torgau Elbe River and hence are not expected to arrive and Wittenberg are given in Table 1. Dikes run- at the storage area. ning along both sides of most of the Middle Elbe The national flood forecasting system pro- River are designed for a 100 year flood plus one vides water level information with lead-times of metre freeboard. Accordingly, the proposed 48 hours plus another two to three days trend at emergency storage area is to be designed for re- the river stretch of interest. The accuracy of the ducing flood peaks of 100 years return period or flood forecast model is given with a standard er- larger as the dikes are expected to withstand ror of 5 cm (BfG, 2006). Unless unexpected floods of smaller magnitude. However, very large events such as dike failures occur just upstream floods in the range of 200 years recurrence time of the storage area, the flood forecast informa- or larger are very likely to cause dike overtopping tion is accurate enough and gives sufficient time or dike failure at upstream locations along the to enable an optimised operation of the control structures.

Tab. 1: Discharge statistics for the gauges Torgau and Wittenberg (LHW, 2003). Lower discharge values recorded at the downstream gauge Wittenberg are due to several dike failures that occurred between the two gauges during the flood in 2002. Gauge (Elbe-km) Mean annual Mean annual Maximum discharge Series discharge peak discharge recorded [m³/s] [m³/s] [m³/s] (date)

Torgau (154.2) 344 1420 4420 (18.08.2002) 1936-2003

Wittenberg (214.1) 369 1410 4110 (18.08.2002) 1936-2003

30 Hydrodynamic simulation of the operational management …

Fig 3: Longitudinal cross-section along the modelled river stretch showing Elbe river bottom, and proposed polder dikes as well as control structures

Fig 4: Cross-sectional view of the southern polder basin

Chapter III 31

It may be noted that the emergency storage well as other processes and measures that lead to area under study is a potential detention site and a modification of the river cross section it should hence, at present neither the dikes surrounding be ensured that up-to-date cross section data is the polder basins nor any control structures exist. used. Topographic data and information on control The boundary condition at the upstream end structures and polder dikes were provided by the of the reach at 175.00 Elbe-km is a discharge hy- local water authorities according to the current drograph of the Torgau gauging station. Al- planning stage. though this gauge is located approximately 20 km Currently about 90 % of the area is under in- upstream of the upper model boundary, utilisa- tensive agricultural use. The rest is taken up by tion of these discharge data is justified as there watercourses and forests, most of which are un- are only minor tributaries to the Elbe River be- der nature protection. It is expected that the land tween Torgau and the modelled river stretch. The will retain its original purpose after designation as downstream boundary condition at 193.60 Elbe- emergency storage area. km is provided as a rating curve. The emergency storage area is schematised in the model by two storage cells each representing 3 HYDRODYNAMIC MODEL SETUP one polder basin. The storage cells are described by their area-elevation curves. The curves are de- Models of different complexities may be ap- rived from a high-resolution Digital Elevation plied for the hydrodynamic simulation of flood- Model that was obtained from airborne LiDAR plains and storage areas (Bates et al., 2005, Hes- survey. The gates are implemented as control selink et al., 2003). An appropriate modelling structures that operate due to pre-set conditions approach should be chosen depending on the ob- according to the control strategies described in jective of the study, the available data and the section 4.3. amount of accuracy required (Bates and De Roo, The model is carefully calibrated and validated 2000). A two-dimensional or quasi-two- against water levels recorded at the gauging sta- dimensional hydrodynamic model of the storage tion at 184.4 Elbe-km. Because of the different area is essential if the propagation of the inunda- nature of bed materials two hydraulic roughness tion flow in the storage area as prerequisite for classes are distinguished, one for the main chan- the simulation of erosion and sedimentation nel and one for the floodplain. Initially, the processes or the subsequent estimation of flood roughness information in the form of Manning’s damage is of interest (Huang et al., 2007, Chatter- n values is taken from the literature (Chow, 1959) jee et al., 2007). However, if one is only inter- and similar studies (Horritt and Bates, 2002). A ested in studying the effectiveness of the storage two stage calibration and validation procedure is area in capping the peak discharge in the river, a adopted in order to find the appropriate rough- simple one-dimensional model for the river cou- ness coefficients. In the first stage the coefficient pled with a storage cell may be sufficient. The lat- for the river bed is identified by considering two ter approach was chosen for this study where the smaller flood events during which the water did simulation of the filling and emptying process of not spill over to the floodplains. In the second the emergency storage area is made by the practi- stage the roughness coefficient for the floodplain cal application of the modelling package MIKE is found by use of two larger flood events which 11 developed at the Danish Hydraulic Institute, also affected the floodplains. In this stage the co- Denmark (DHI, 1997). MIKE 11 has been ap- efficient for the main river is kept to the value plied in several similar studies (Faganello and At- that was identified to fit the data best during the tewill, 2005, Hammersmark, 2002). It uses an im- first stage. Finally, Manning’s n values of 0.038 plicit finite difference scheme for computation of for the river channel and 0.050 for the floodplain unsteady flow based on the vertically integrated were found to fit the observation data best on the equations of conservation of continuity and mo- basis of a goodness of fit criterion. Due to lack of mentum (the Saint Venant equations). calibration data for the emergency storage area, MIKE 11 is set up to represent a 18.6 km the floodplain roughness value obtained in the reach of the Elbe River, which is described by a calibration process is also applied to the polder series of 34 cross sections that stem from recent basins. This assumption seems acceptable, since sonar measurements and airborne laser altimetry land use in both the floodplains and the deten- (see Figure 5). In view of future bed erosion as tion areas are comparable. For a detailed descrip- 32 Hydrodynamic simulation of the operational management … tion of the model setup, calibration process and sensitivity analysis see Chatterjee et al. (2008).

Fig. 5: Cross-sectional view of the southern polder basin

flood scenarios are found by analysing peak flows observed at Torgau gauge. 4 CONSIDERED FLOOD SCENARIOS For the management of the emergency storage AND CONTROL STRATEGIES area different control strategies may be adopted with respect to timing, sequence and number of A primary objective of this study is to investigate operations. Out of these two most promis- gate the effect of magnitude and shape of a flood ing control strategies are chosen for further in- wave on the peak attenuation. Rather than study- vestigation. ing a single past flood event with a specific magnitude and hydrograph shape, in the present study flood scenarios were derived in such a way 4.1 Flood frequency analysis that they cover a broad range of potential flood An at-site flood frequency analysis is carried situations to be expected at the storage area. The out in order to relate the magnitude of extreme Chapter III 33 flood events to their frequency of occurrence. The hydrograph shapes have been obtained by The analysis is based on the annual maximum se- analysing the ten largest floods in the 70 year dis- ries (AMS) of the 1936-2005 discharge record at charge record at the Torgau gauging station. the Torgau gauging station. Torgau gauge is situ- Among them the flood event of summer 2002 ated approximately 20 km upstream of the stor- was the only one exceeding the discharge corre- age area along the Elbe River. sponding to a return period of 100 years. In a Preliminary tests for independence and ho- non-dimensional analysis the peaks of these hy- mogeneity as well as tests for outliers are carried drographs were scaled to a uniform ordinate out. The best fit distribution was identified on value of 1. The scenario hydrographs were then the basis of the L-moment ratio diagram and the constructed by selecting the steepest and widest Z-dist statistic criteria (Hosing and Wallis, 1997). hydrographs considering the top portion of the Among several 2- and 3- parameter distribution flood wave that is relevant for the diversion of functions the Generalised Logistic Distribution water into the storage area (above approximately was found to fit the data best. Based on the se- 3500 m³/s, see Table 2). Subsequently, each dis- lected distribution function, the 100 and 200 year charge ordinate was multiplied by an amplifier, return period floods are found to have a dis- which is the design peak corresponding to the charge of 4022 and 4775 m³/s, respectively. The 100 and 200 year return period. Besides the in- maximum discharge observed at Torgau station crease in peak discharge the described transfor- on 18th August 2002 was 4420 m³/s (Table 1) mation lead to an enhanced flood wave volume which corresponds to a return period of ap- and duration in the top portion of the hydro- proximately 180 years. graph that is relevant for peak capping. Figure 6 shows all four considered flood scenarios. Several two or multiple peaked floods can be 4.2 Definition of flood scenarios found in the Torgau discharge record. This is According to the water authority the flood also the case with the selected steep flood hydro- storage area will be utilised to reduce large flood graph where a second flood peak follows the lar- peaks ranging between 100 and about 200 years ger peak. However, the second peak is much recurrence time. This range is a result of the cur- smaller than the discharge threshold relevant for rent dike heights as described above. Conse- water diversion into the storage area and hence, quently, flood magnitudes of 100 and 200 years does not have an effect on the operation of the return period are considered in this investigation. storage area.

Fig. 6: Hydrographs of investigated flood scenarios as derived from the discharge record at the Torgau gauge. Hydrographs of the flood events of April 1941 (steep) and March/April 1988 (wide) were rescaled to the peak discharge corresponding to the 100 and 200 return period, respectively

which have an effect on the flood peak attenuation as well as on the flooding dynamics within 4.3 Definition of control strategies the storage area. The type of control strategies refers to the Two control strategies are considered in the timing, sequence and number of gate operations, study as schematised in Figure 7. Flow direction 34 Hydrodynamic simulation of the operational management … of the Elbe River is from left to right. The cur- gate closure process. The threshold value is de- rent state of opening of the three gates is marked termined for each individual flood hydrograph. by black and white arrows. As can be seen, the To complete the filling process, the CG and south basin has a higher design water level than afterwards the SG are closed as soon as the north the north basin. Consequently, both polder ba- basin and the south basin, respectively, reach sins cannot be completely filled up solely through their full capacity (3, 4). the north gate. On the contrary, in control strategy II the In the simulation runs the gates are operated emergency storage area is filled through the NG according to pre-defined event-specific condi- as well as the SG. Following the same initial gate tions of Elbe discharge and polder water level. In status (1), a pre-defined number of NG parts is control strategy I the north gate (NG) and the opened first upon reaching a certain Elbe thresh- south gate (SG) are initially closed whereas the old discharge and hence the filling process starts sluice gate connecting both polder basins (CG) is from the low lying areas (2). The number of gate open (1), as shown in Figure 7. When a certain parts which open depends on the required dis- Elbe threshold discharge is reached during a large charge capacity of the gate, which will be high for flood event the first part of the SG is opened rather steep flood hydrographs and lower for leading to a momentary drop followed by an in- wide hydrographs. After opening the SG upon crease in Elbe water levels. The remaining SG reaching a threshold Elbe discharge, the storage parts are then opened every time the Elbe area is filled from both sides (3). This process threshold level is reached (2). continues until the NG is closed because the wa- The selection of this threshold value corre- ter flow through the gate reverses due to decreas- sponding to the maximum achievable attenuation ing Elbe water levels compared to increasing wa- is governed by the available volume in the storage ter levels in the storage area (4). Similar to control area, the discharge capacity of the inlet gates and strategy I the CG and the SG are closed as soon their opening/closure duration. It was assumed as the design water levels in the polder basins are that the gates take 30 minutes for a complete reached (5, 6). opening process, which is a realistic duration for In both control strategies the emptying proc- such type of gates according to the water authori- ess starts as soon as the Elbe water level falls be- ties. Once the process is started it continues until low a level that allows for safe discharge at the the gate is fully open. The same applies to the downstream river reaches. The storage area is drained out through NG and SG.

Fig. 7: Schematic cross-sectional view of the polder basins with gate operations for filling process according to control strategy I (left) and II (right)

Chapter III 35

The results show that the water level attenuation in the Elbe River highly depends on the 5 RESULTS steepness of the flood hydrograph. Water level reductions for both flood scenarios differ by Figures 8-11 show selected results obtained more than 10 cm. The filling time amounts to from simulation runs with varying hydrograph about 3 days in case of the wide hydrograph (Fig- shape, return period and control strategy, ure 9). The magnitude of the flood, however, whereas Table 2 summarises the peak attenuation does not influence the peak attenuation amount for each case. The resulting water level lowering provided that they have the same hydrograph in the river given in Table 2 corresponds to the shape (Table 2). maximum difference between the peaks of the A comparison of the two control strategies unaffected hydrograph and simulated attenuated shows only slightly higher water level attenuation hydrograph, while the highest water levels do not for control strategy II in case of the steep hydro- necessarily occur at the same time. The maximum graph (Figure 10). For the complete filling of the possible water level reduction refers to the ideal polder basins during very steep flood events it is case of a horizontal flood peak capping when us- advisable to increase the inlet discharge by addi- ing the full capacity of the storage area. tional utilisation of the northern control structure Scenario results are discussed for the case of as done in strategy II. A volume of 2.4 million m³ the steep hydrograph with a return period of 200 flows through the north gate before the filling is years when control strategy I is applied with the completed through the southern gate. Due to the gates opening sequentially (Figure 8). The other lower elevation heights of the northern basin this simulations are analogous. When a threshold dis- strategy results in a pre-filling of this basin. Strat- charge of 4200 m³/s is reached, the first south egy II may therefore also be advantageous in gate part is opened. Consequently flow through view of erosion mitigation. However, a detailed the south and the connecting gate starts and the study of the flow velocities is required to confirm discharge in the Elbe River drops and subse- this aspect and is a subject of future work. In quently increases again with the rising limb of the view of the operational use, control strategy I flood wave. A volume of approximately 0.1 mil- may be easier to adopt because the number of lion m³ flows through the south gate in the first gate operations is less. 30 minutes until the first gate part is fully open. A complete opening of the inlet south gate at The other south gate parts are sequentially once rather than opening the gate parts sequen- opened every time the threshold discharge is tially causes a sharp drop in the Elbe discharge, reached. After the south gate is completely open which is not desirable (Figure 11). The water the discharge through the gate amounts to ap- level reduction obtained here is lower (18 cm) proximately 450 m³/s, whereas the discharge compared to the strategy of sequential opening of through the connecting gate is approximately 220 the gate parts (21 cm) for the case of the steep m³/s. The filling process continues for about 1.5 hydrograph when control strategy I is applied. days until the design water levels in the polder Hence, the simulation runs for all scenarios were basins are reached and consequently the gates are done applying the favourable sequential opening closed. Subsequently the water levels in the pol- of gate parts. There are also real operation exam- der basin remain constant for about 8 days until ples for the sequential inlet of flood water (see the emptying process through south and north emergency storage area at the River Rhine in gates starts. When the flow direction at the south Bettmann and Bauer, 2005). gate reverses, the gate is closed and subsequently As mentioned above the investigated storage the polder is drained through the north gate. A area is not yet constructed and hence only cali- water level reduction of 21 cm just downstream bration data for the river is available. Due to the of the storage area (at Elbe River 189.6 km) cor- lack of observation data for the storage area the responding to a discharge of approximately 450 simulation results are compared with a similar m³/s was simulated for the described scenario. study. In IWK (2003) the peak attenuation effect This equals a relative discharge attenuation of for several proposed flood storage areas along 9.5%. In a similar study on a system of two pol- the Middle Elbe River was simulated considering ders at the Danube River in southern Germany a floods with peak discharges ranging between relative peak reduction of 11% was simulated 4000 m³/s and 5000 m³/s. For the same storage (Fischer et al., 2005). area as investigated in the present study a peak 36 Hydrodynamic simulation of the operational management … reduction between 262 m³/s (14 cm) and 497 risk of dike failures due to overtopping, piping or m³/s (23 cm) at the gauge Wittenberg was simu- saturation downstream along the river and hence lated. These results are very similar to the range decrease the risk of extensive inundations. The of water level reduction obtained in the present investigated storage area is part of a series of de- study. tention areas that is currently planned at the In both studies the calculated peak attenuation Middle Elbe River. Once they are in operation seems rather low. However, a water level de- each of the areas will contribute to a larger over- crease in the range of 10 to 20 cm may reduce the all water level reduction.

Fig. 8a: Gate levels: steep hydrograph, HQ200, control Fig. 9: Simulated discharge and water levels: wide hydro- strategy I, gate opening in parts (NG north gate, CG con- graph, HQ200, control strategy I, gate opening in parts necting gate, SG I, II, III, IV south gate parts)

Fig. 8b: Gate discharge: steep hydrograph, HQ200, control Fig. 10: Simulated discharge and water levels: steep hydro- strategy I, gate opening in parts graph, HQ200, control strategy II, gate opening in parts

Fig. 8c: Simulated discharge and water levels: steep hydro- Fig. 11: Simulated discharge and water levels: steep hydrograph, HQ200, control strategy I, gate opening in parts graph, HQ200, control strategy I, gate opening in one part, that is the entire 25m gate width opens together Chapter III 37

Tab. 2: Maximum possible water level reduction, applied Elbe threshold discharge, simulated Elbe water level reductions and number of gate parts used during filling process for the investigated flood scenarios and control strategies Steep Wide Steep Wide hydrograph, hydrograph, hydrograph, hydrograph, HQ200 HQ200 HQ100 HQ100 Peak discharge (m³/s) 4775 4775 4022 4022 Maximum possible water level 24 11 23 11 reduction (cm) Control Strategy I (Only South Gates) Threshold discharge (m³/s) 4200 4575 3525 3825 Water level reduction (cm) 21 9 19 9 Number of gate parts used 4 SG 2 SG 4 SG 2 SG Control Strategy II (Combination of North & South Gates) Threshold discharge (m³/s) 4225 4575 3550 3825 Water level reduction (cm) 22 9 21 9 Number of gate parts used 2 NG + 4 SG 2 NG + 1SG 2 NG + 4 SG 2 NG + 2 SG

area in the short time a steep hydrograph passes by, it is advisable to increase the inflow capacity 6 CONCLUSIONS AND DISCUSSION by utilising both control structures during the filling process. However, differences in peak cap- The present study shows that from a hydraulic ping between both control strategies are very point of view the proposed detention site seems small. suitable for an effective peak reduction. This is For the operational use of this specific emer- mainly due to the large volume and the good gency storage area further detailed investigations drainage conditions of the investigated storage need to be carried out, in particular for the design area. A local water level attenuation between 9 of the hydraulic structures and their optimised and 22 cm was simulated. operation. The study also shows that the peak reduction Based on the experience gained in the present strongly depends on the steepness of the flood study general conclusions about the management hydrograph. Hence, the river threshold discharge of controlled storage areas can be drawn. at which to open the control gate should be cho- Flood storage in detention basins is most ef- sen depending on the forecasted hydrograph fective if it is controlled by adjustable gates. Di- shape. Unless there are large dike failures or viding the inlet gate into separately operable parts gauge malfunctions upstream of the storage area, has proved to be a promising strategy for a con- the forecast information provided by the national trolled water inflow. It also allows for the utilisa- flood forecasting system should be accurate and tion of the storage area in case of a gate malfunc- early enough to determine the required threshold tion. The existence of two or more separate discharge in advance and hence enable an opti- polder basins can contribute to the flexibility of mised operation of the studied polder system. the polder operation because it enables the appli- A specific characteristic of the investigated cation of different control strategies according to storage area is the existence of two polder basins, the forecasted flood wave. each equipped with control structures. This facili- When planning further detention sites as cur- tates an event-adapted management with differ- rently done at the middle Elbe River it is impor- ent control strategies. It was seen that the peak tant to study the effect of different hydrograph reduction effect for steep hydrographs is signifi- shapes on the peak attenuation as this affects the cantly larger than for relatively wide flood hydro- damage to be expected and hence the overall graphs. In order to completely fill the storage cost-benefit analysis for a certain detention site. 38 Hydrodynamic simulation of the operational management …

ACKNOWLEDGEMENTS FLOODsite Consortium is liable for any use of the information in this paper. The research was jointly funded by the Alex- Data were kindly provided by the following ander-von-Humboldt Foundation Fellowship authorities: Landesbetrieb für Hochwasserschutz Program and the Sixth Framework Program of und Wasserwirtschaft Sachsen-Anhalt, Wasser- the European Commission (FLOODsite project, und Schifffahrtsamt Dresden and Landesvermes- EC Contract number: GOCE-CT-2004-505420). sungsamt Sachsen-Anhalt. The authors are also This paper reflects the authors’ views and not grateful to the Danish Hydraulic Institute (DHI), those of the European Community. Neither the Denmark for providing an evaluation copy of the European Community nor any member of the MIKE 11 software.

Chapter IV

Comparison of hydrodynamic models of different complexities to model floods with emergency storage areas

ABSTRACT: A flood emergency storage area (polder) is used to reduce the flood peak in the main river and hence, protect the downstream areas from getting inundated. In this study, the effectiveness of a proposed flood emergency storage area at the middle Elbe River, Germany in reducing the flood peaks is investigated using hydrodynamic modeling. The flow to the polders is controlled by adjustable gates. The extreme flood event of August 2002 is used for the study. A fully hydrodynamic 1D model and a coupled 1D-2D model are applied to simulate the flooding and emptying processes in the polders and flow in the Elbe River. The results obtained from the 1D and 1D-2D models are compared with respect to the peak water level reductions in the Elbe River and flow processes in the polders during their filling and emptying. The computational time, storage space requirements and modeling effort for the two models are also compared. It is concluded that a 1D model may be used to study the water level and discharge reductions in the main river while a 1D-2D model may be used when the study of flow dynamics in the polder is of particular interest. Further, a detailed sensitivity analysis of the 1D and 1D-2D models is carried out with respect to Manning’s n values, DEMs of different resolutions, number of cross- sections used and the gate opening time as well as gate opening/closing duration.

Published as: Chatterjee C, Förster S and Bronstert A. 2008. Comparison of Hydrodynamic Models of Different Com- plexities to Model Floods with Emergency Storage Areas. Hydrological Processes. Published online, DOI: 10.1002/hyp.7079. 40 Comparison of hydrodynamic models of different complexities …

ware for 2D modeling are FLO 2D (O’Brien, 2006), RMA2 (King et al., 2001), MIKE-21 1 INTRODUCTION (DHI, 2000), DELFT-FLS (Hesselink et al., 2003), DELFT-3D (Stelling and Duinmeijer, Flood storage areas form part of the flood 2003) and TELEMAC-2D (Horritt and Bates, management strategy at many lowland rivers. 2001). They are used to temporarily store excess The 1D models fail to provide information floodwater in order to reduce peak flood flows on the flow field while the 2D models require downstream. In the last few years various stud- substantial computer time; hence, attempts ies have been conducted to predict the flood have been made to couple 1D river flow mod- peak reduction in the main river and to simulate els with 2D floodplain flow models. In the the inundation process in the flood storage ar- coupled 1D-2D models, the flow in the main eas. These studies represent a special case of river channel is simulated using the 1D equa- hydrodynamic floodplain modeling because of tions, while the 2D equations are solved for the the controlled manner of detention and release water spilling over the banks to the floodplains. of flood water. Flow may be controlled by The link between the two kinds of flow is usu- spillways, adjustable inlet and outlet structures ally done by a mass conservation equation. or engineered dike breaches. Dhondia and Stelling (2002) describe the 1D- Various researchers have used the hydrody- 2D model SOBEK (Rural/Urban) developed namic modeling approach to simulate flood in- by the laboratory at Delft Hydraulics. The undation in the floodplains (Werner, 2004; MIKE-21 model has been dynamically linked to Bates et al., 2005). The hydrodynamic modeling the MIKE-11 model, into a single package approach of flood inundation simulation essen- called MIKE FLOOD developed at the Danish tially involves the solution of one dimensional Hydraulic Institute (Rungo and Olesen, 2003). and two dimensional Saint Venant equations Hydrodynamic models of different complex- using numerical methods. Various numerical ity have been used to simulate flow situations models have been developed for flood plain de- with flood storage areas. 1D models for the lineation/flood inundation and flow simulation. river coupled with storage cells that represent These numerical models essentially involve the polders have been used to simulate the peak solving the governing equations for flow in riv- capping effect. The storage cells are usually ers and floodplains using certain computational characterised by the relationship between vol- algorithms. Based on the approximations used, ume or area as a function of elevation or by a the numerical models are categorised into (a) series of cross sections (Minh Thu, 2002; one-dimensional (1D) models, (b) two- Kúzniar et al., 2002; Faganello and Attewill, dimensional (2D) models, and (c) one- 2005). Also, there are a few studies applying dimensional river flow models coupled with combined 1D-2D model, where the flow in the two-dimensional floodplain flow (1D-2D) river is solved in 1D whereas the flow in the models. storage area is simulated using a 2D approach Various software like HEC-RAS (HEC (Baptist et al., 2006). Further, since full 2D or River Analysis System) from the U.S. Army combined 1D-2D models are generally compu- Corps of Engineer’s Hydrologic Engineering tationally very extensive, quasi 2D model ap- Center (HEC, 2002), U.S. National Weather proaches have been applied in order to give a Service’s (NWS) DWOPER and FLDWAV simplified 2D representation of the storage area (Fread et al., 1998), MIKE11 developed at the combined with a fast computation process Danish Hydraulic Institute, Denmark (DHI, (Aureli et al., 2005; Huang et al., 2007). 1997), SOBEK-1D developed at the Delft Hy- As far as floodplain inundation is concerned, draulics, Delft (Werner, 2001) etc. have been several studies have been carried out to com- used extensively for dynamic 1D flow simula- pare the predictive performances of hydrody- tion in rivers. The 1D models though simple to namic models of different complexities (Horritt use and provide information on bulk flow char- and Bates, 2002; Tayefi et al., 2007). However, acteristics, fail to provide detailed information there has been no such detailed comparative regarding the flow field. Hence, attempts have study for simulating floods with emergency been made to model the 2D nature of flood- storage areas. In this study, a comparison of 1D plain flow. Some of the most widely used soft- MIKE11 and 1D-2D MIKEFLOOD models Chapter IV 41 to simulate flows for a proposed flood emer- storage area is designed for reducing flood gency storage area at the middle Elbe River, peaks of not less than 100 years return period. Germany is carried out. It has to be emphasised that the emergency storage area is a potential retention site and hence, at present neither the dikes surrounding 2 STUDY AREA AND DATA USED the polder basins nor any control structures exist. Topographic data, river flow data and in- The proposed emergency storage area is lo- formation on control structures and polder cated in the lowland area at the Middle Elbe dikes were provided by the local water authori- River, Germany (Figure 1). It extends 7 km ties according to the current planning stage. along the right bank of the Elbe River. The In this study, the flood event of the period overall area amounts to 17 km² with a maxi- 5th Aug. to 17th Sep., 2002, i.e. a duration of 44 mum storage capacity of approximately days is considered to study the effectiveness of 40 million m³. the proposed polder. For this flood event, the The storage area is divided into a northern peak discharge is 4420 m3/s which occurred on and a southern polder basin by an already exist- 18th Aug, 2002 at the gauge of Torgau. Besides, ing dike road running through the area. Both four other flood events are used for model cali- basins are connected by a sluice gate of 50 m bration and validation as described later. width and 2 m depth. This gate is termed as the Currently about 90 % of the area is under in- connecting gate. The filling and emptying proc- tensive agricultural use. The rest is taken up by ess of the storage area is to be controlled by watercourses and forests, most of which are two adjustable overflow weirs of 25 m width under nature protection. It is expected that the each. The gate for the north polder is termed as land will retain its original purpose after desig- the north gate while the gate for the south pol- nation as emergency storage area. der is termed as the south gate. The emergency

Fig. 1: Map of the proposed flood emergency storage area at the Middle Elbe River, Germany 42 Comparison of hydrodynamic models of different complexities …

sisting of a three point (upstream cross-section, structure and downstream cross-section) 3 METHODOLOGY MIKE11 branch whose ends are linked to MIKE21 cells, and (iv) zero flow link specified The one-dimensional model MIKE 11 and to a MIKE21 cell will have zero flow passing the coupled one-/two-dimensional model across the cell (DHI, 2004). For this study, the MIKEFLOOD are applied to simulate the standard link is the only relevant link that can flooding and emptying processes in the polders be used and hence, it is selected. and flow in the Elbe River. The governing flow equations of MIKE 11 are one dimensional and are of shallow water types which are the modi- 3.1 One-dimensional model setup fications of Saint Venant equations (DHI, 1997; The 1D model MIKE 11 is set up to repre- DHI, 2000). MIKEFLOOD integrates topog- sent a 18.6 km reach of the Elbe River (Fig- raphic data of the one-dimensional MIKE 11 ure 2), which is described by a series of 34 cross river network with the two dimensional MIKE sections. These cross-section data are provided 21 floodplain (bathymetry data) through four by the local water authority. The cross-section different linkages i.e. (i) standard link where one data available downstream of Elbe 187 km are or more MIKE21 cells are linked to the end of not only very sparse but these are available for a MIKE11 branch, (ii) lateral link where a string the main channel only. Hence, these cross- of MIKE21 cells are laterally linked to a speci- sections were extended to the dikes using eleva- fied reach of MIKE11, (iii) structure link con- tion data from airborne laser altimetry.

Fig. 2: MIKE 11 model layout for the proposed flood emergency storage area Chapter IV 43

The boundary condition at the upstream end these are used for calibration and validation for of the reach at 175.0 Elbe-km is a discharge hy- the main channel. Initially, the roughness in- drograph of the Torgau gauging station. Al- formation in the form of Manning’s n values is though this gauge is located approximately 20 taken from the literature (Chow 1959) and simi- km upstream of the upper model boundary, lar studies (Horritt and Bates 2002) which are utilisation of these discharge data is justified as then modified during the calibration process. there are only minor tributaries to the Elbe Two goodness of fit criteria are used to River between Torgau and the modelled river compare the simulated water levels with the ob- stretch. The downstream boundary condition at served values. These are (i) the Nash-Sutcliffe 193.6 Elbe-km is provided as a rating curve. coefficient (Ens) and (ii) the index of agreement The emergency storage area is schematised (d), which are as follows: in the model by two storage cells each representing one polder basin. The storage cells are 2  QQ so  described by their area-elevation curves (Fig- Ens=1 2 (1) ure 2). The curves are derived from a high-  QQ avo resolution Digital Elevation Model that was obtained from airborne LiDAR survey. The gates 2   QQ so  are implemented as control structures that op- d=1 2 (2) erate due to certain pre-set conditions as men- avo  QQQQ avs tioned later. Where, Ens=Modeling efficiency, Qo= Ob- 3 3.1.1 Calibration and validation of model served discharge (m /s), Qs = Simulated dis- 3 Calibration of the hydrodynamic models for charge (m /s), Qav= Mean of the observed dis- emergency flood storage areas is difficult as charge (m3/s) and d= Index of agreement. emergency storage areas are usually designed to be used for flood peak capping during rare 3.2 One-/ two-dimensional model setup flood events. The storage area may have never been in operation before and hence, observa- In the MIKEFLOOD model layout, the tion data for calibration purpose may not be Elbe River and the three gates (inlet or south available. The same is true for the present study gate, connecting gate and outlet gate) are repre- which investigates a proposed flood storage sented in the 1D model MIKE11. The polders area that is not yet constructed. However, the are represented in the form of a DEM in the MIKE11 model is calibrated and validated for 2D model MIKE21. Both the 1D MIKE11 and the flow in the Elbe River and its floodplain 2D MIKE21 models are dynamically coupled within the embankments. by standard links. This coupled model is run The MIKE11 model is calibrated and vali- with the same boundary conditions, flood sce- dated against water levels recorded at the Mau- nario (i.e. for the August 2002 flood event) and ken gauging station at 184.4 Elbe-km. Because gate operation as the 1D model. Thus, the es- of the different nature of bed materials two hy- sential difference between the 1D MIKE11 draulic roughness classes are distinguished, one setup and MIKEFLOOD setup is in the repre- for the main channel and one for the adjacent sentation of the polders. floodplain. In order to identify the two rough- In this study, three different DEMs are used ness coefficients, a two stage calibration and which are obtained by resampling the LIDAR validation procedure is adopted. In the first DEM to grid sizes of 8 m, 25 m and 50 m. Ini- stage, the roughness coefficient for the main tially, the LIDAR DEM is processed to remove channel is identified and in the second stage the non-permanent objects, such as dung hills or roughness coefficient for the floodplain is iden- vehicles, and to correctly represent line struc- tified. Four flood events during the period 1st tures, such as dikes and ditches. While generat- October to 30th Nov 1999, 1st January to 31st ing a DEM by interpolation of point data ob- March 2002, 1st August to 27th September 2004 tained by laser scanning, there is a risk that line and 1st March to 30th June, 2005 are selected for structures are not continuous in the DEM. the process of calibration and validation. Of Thus, the following procedure is used for cor- these, water does not spill over to the flood- rect representation of line structures: (i) line plains for the first and third events and hence, structures are digitised as line objects using to- 44 Comparison of hydrodynamic models of different complexities … pographic maps, (ii) height information is at- of the MIKE11 model setup to the Manning’s tached to the digitised line objects, (iii) a large n values is carried out by including the polders. number of points are generated along the line Again, the same two cases of n values stated objects and (iv) the DEM is generated by inter- above are considered (Table 1). polating point data that were collected by the The sensitivity analysis of the laser scanner and generated from the line ob- MIKEFLOOD setup to the Manning’s n values jects using GIS. As the laser scanner only col- is also carried out. In this case the n values for lects water surface elevation, bottom height of the river as well as the river floodplain are kept ditches was obtained by terrestrial measure- at their calibrated values while the n values for ments and included in the DEM generation the polders are varied i.e. increased and de- procedure. While aggregating LIDAR data to creased by 5% (Table 1). other grid sizes, the DEM gets “smoother”, i.e. dikes have lower elevation and ditches become Table 1: Range of Manning’s n values for different land- shallower. In order to preserve the flooding use class considered in sensitivity analysis characteristics, post-processing is done in the Class Manning’s n* aggregated DEM by converting the line objects (with correct height information) to grid objects River channel 0.0361 – 0.0399 (0.038) of the same grid size as the aggregated DEM. River floodplain 0.0475 – 0.0525 (0.050) Subsequently, the corresponding grid cells in Polder 0.0475 – 0.0525 (0.050) the aggregated DEM are replaced by grid cells *Calibrated values are indicated in brackets of the line object. While aggregating the DEM to grid sizes of 8m, 25m and 50m, it is ensured 3.3.2 DEM’s of different resolutions that water does not spill over the dikes (Elbe dike and polder dikes) by setting the dike cells The sensitivity of the 1D MIKE11 model to to their true elevations. However, the correct the use of different DEM resolutions is studied. depths of the ditches inside the polders are not Two DEMs of horizontal resolution 8 m and included in the aggregated DEMs as these 50 m are used to derive the area-elevation ditches are very narrow (about 3 to 5 m wide) curves. The MIKE11 model is simulated for and hence, would be overrepresented when the August 2002 flood event with the area- converting them into grid sizes of 8 m, 25 m elevation curves derived from the two different and 50 m. DEMs. The sensitivity of the MIKEFLOOD model to the use of different DEM resolutions is also 3.3 Sensitivity analysis studied. The sensitivity analysis is carried out A detailed sensitivity analysis is carried out considering three DEM’s of different horizon- for the different hydrodynamic models with re- tal resolutions for the polders viz. 8 m, 25 m spect to a number of input parameters like (i) and 50 m. The sensitivity of the use of these Manning’s n values, (ii) DEM’s of different different DEM’s to the water level and dis- resolutions, (iii) number of cross-sections used charge reduction in the Elbe River as well as and (iv) gate opening time and opening/closing the flow dynamics in the polders is studied. The duration. The conditions under which the sensi- flood inundation extent and depth in the pol- tivity analyses are carried out for each of these ders at a particular instant of time for the dif- input parameters is presented below. ferent DEM’s is also studied.

3.3.1 Manning’s n 3.3.3 Number of cross-sections used First the sensitivity analysis of the MIKE11 The sensitivity of the 1D MIKE11 model to model setup for only the Elbe River (i.e. with- different number of cross-sections used is also out the polders) is carried out with respect to studied. In this case also, only the Elbe River is Manning’s n values. For this purpose, two cases modeled and the polders are not considered. As of Manning’s n values are considered. In the stated earlier, a total of 34 cross-sections are first case, the n values are decreased by 5% used to define the Elbe River in the MIKE11 from the mean/calibrated values while in the model (Figure 2). These cross-sections are in second case, the n values are increased by 5% general 400 m to 800 m apart but the cross- (Table 1). Subsequently, the sensitivity analysis section spacing is more in the downstream side Chapter IV 45 with a maximum spacing of 2.4 km between 4 RESULTS AND DISCUSSIONS Elbe River 189.6 km and 192 km. In order to study the sensitivity of the results to the number of cross-sections used, two different cases 4.1 Calibration and validation of the one- are considered in which different number of dimensional model cross-sections are used: Case-I: Only 20 out of 34 cross-sections are Table 2 shows the performance indices for used, i.e. 14 cross-sections (nos. 3, 5, 7, 9, 11, different trial values of Manning’s n for 13, 15, 17, 19, 21, 23, 25, 27 and 29 (Figure 2)) MIKE11 simulated water levels at the Mauken are removed. Here, the 14 cross-sections which gauging site during calibration and validation are removed are in the stretch of the Elbe River for the main channel only. The Ens and d values 175 km to 187.6 km. In this stretch of the river, are found to be the highest for n equal to 0.038 the cross-section spacing varies from 400 to during calibration. Using this n the Ens and d 800 m, i.e. they are closely spaced. Thus, the re- values are also found to be very high during sults obtained from the removal of these cross- validation. Hence, the Manning’s n value of sections would indicate the closeness at which 0.038 is chosen for the main channel. the cross-sections are to be provided. Case-II: Again another set of 20 cross- Table 2: Performance indices for MIKE11 simulated wa- sections are used, i.e. a different set of 14 cross- ter levels at the Mauken gauging site during calibration sections are removed (nos. 2, 4, 6, 8, 10, 12, 14, and validation (for the main channel only) 16, 18, 20, 22, 24, 26 and 28 (Figure 2)) in the Calibration Validation same stretch of the Elbe River 175 km to 187.6 Events  Oct-Nov, 1999 Aug-Sep, 2004 km. Manning’s Ens dEns d n (for river) 3.3.4 Gate opening time and opening/closing duration 0.037 0.662 0.931 - - The sensitivity of the 1D MIKE11 model to 0.038 0.925 0.984 0.92 0.983 the time of opening of the south gate during 0.039 0.844 0.967 - - the polder filling process is studied. For this the following two gate opening times are consid- Table 3: Performance indices for MIKE11 simulated wa- ered for the south gate: ter levels at the Mauken gauging site during calibration and validation (for the floodplains) Case-I. 6 hour ahead of the actual opening Calibration Validation time. Events  Jan-Mar, 2002 Mar-Jun, 2005 Case-II. 6 hour after the actual opening time. Manning’s n Ens dEns d For For The sensitivity of the 1D MIKE11 model to river floodplain different gate opening and closing duration dur- 0.038 0.035 0.976 0.724 - - ing the polder filling process is also studied. In 0.04 0.98 0.728 - - this study all the gates (i.e. the south and north 0.045 0.981 0.732 0.979 0.584 as well as the connecting gates) open or close in 0.05 0.979 0.736 0.982 0.594 a span of 30 min (based on information from local water authority). In order to study the sen- Different trial values of Manning’s n for the sitivity of the gate opening and closing duration floodplain are chosen keeping the main channel to the water level and discharge reduction in the n value equal to 0.038. Table 3 shows the per- Elbe River, two different gate opening and formance indices for the MIKE11 simulated closing durations are considered: water levels at the Mauken gauging site during calibration and validation for the floodplain. Case-I: All the gates take 5 min to open or For the Jan-Mar, 2002 event, the Ens value is close. found to be the highest for floodplain n value Case-II: All the gates take 60 min to open or equal to 0.045 while the d value is found to be close. the highest for floodplain n value equal to 0.050. But during validation with the Mar-Jun, 2005 event, both the Ens and d values are found to be the highest for floodplain n value equal to 46 Comparison of hydrodynamic models of different complexities …

0.050. Hence, the Manning’s n value of 0.050 is south gate opens when the water level in the chosen for the floodplains. Figure 3 shows a Elbe River reaches a threshold value of 76.94 m comparison of the observed and simulated wa- corresponding to a discharge of 4100 m3/s ter levels at the Mauken gauging site during (Figures 4a and b). At this instant of time a dis- calibration and validation for the floodplains. charge of about 440 m3/s enters through the south gate (Figure 4c) and this results in a sharp (a) reduction in the Elbe discharge and water level (Figure 4a). Subsequently, the connecting gate and the south gates close when the water level reaches the design value in the north and south polders, respectively. The entire filling process takes about 30 hours. After the gates close, the discharge and water level in the Elbe River rise again. The water level reduction at the Elbe River 184.4 km (i.e. at the Mauken gauge) is 25 cm while the corresponding discharge reduction at this point is 310 m3/s (Figure 4a). In order to achieve the maximum water level reduction in the Elbe River for given polder volumes, the discharge in the Elbe River should be as close as possible to a straight line after the (b) filling process starts in the polder. The factors affecting the magnitude of water level reduction in the Elbe River for given polder volumes are (i) time of opening of the gates during the polder filling process, (ii) gate opening/closing duration (iii) gate width or partitioning of the gates and (iv) shape of the flood hydrograph. A detailed investigation on the effect of (i) sequential operation of the north and south gates during the filling process (ii) partitioning of the gates and (iii) shape of the flood hydrograph, on the magnitude of water level reduction in the Elbe River is reported in Förster et al. (2008). In this study, only the south and the connecting gates (and not the north gate) operate during the polder filling process. The gate

Fig. 3: Comparison of observed and simulated water lev- opening/closing durations are 30 min and the els at the Mauken gauging site during (a) Calibration for gate widths are 25 m (based on information col- the flood event of 1st January to 31st March 2002 and (b) lected from local water authority). Further, as Validation for the flood event of 1st March to 30th June, stated earlier, the August 2002 flood hydro- 2005 graph is considered here. Thus, in this study, the time of opening of the south gate during 4.2 One-dimensional model results for the start of the filling process of the polders is flooding and emptying processes in the the only crucial factor for obtaining the maxi- polder mum possible water level reduction in the Elbe River. The time of opening of the south gate Figures 4a to c show the results obtained during the filling process of the polder is de- from MIKE11 simulation for the flooding and cided manually based on a trial and error proc- emptying processes in the polders and flow in ess so as to maximise the water level reduction the Elbe River for the August 2002 flood event. in the Elbe River. Several trial runs are carried The peak discharge for this flood event is 4420 out with different opening times for the south m3/s. Here, the area-elevation curves for the gate (specified in MIKE11 for each trial run) storage areas are derived from a 50 m grid while the connecting and south gates close DEM. It is observed from these figures that the Chapter IV 47 when the design water level is reached in the north and south polders, respectively. For each run the water level reduction in the Elbe River is noted. It is observed that when the south gate (a) is opened corresponding to a water level of 76.94 m at Elbe River chainage 184.4 km (i.e. on 17th Aug, 2002 at 15.40 hrs for the August 2002 flood event), a maximum water level reduction of 25 cm occurs in the Elbe River. The gate operation during the polder emptying process is also decided manually. The objective is to empty the polders as soon as possible. Thus, it is decided to release the water from the polders into the Elbe River as soon as the water level in the river falls below the water level in the polders. Accordingly, it is decided that the emptying process start when the water level in the Elbe River near the south gate falls to 75.64 (b) m (Figure 4a) i.e. 2 days after the filling process ends which allows for a safe release of the flood water. The south gate is opened first followed by the north gate 8 hours later. The connecting gate is opened 7 hours after the north gate is opened (Figure 4b). Immediately after the connecting gate is opened, the south gate is closed as the flow direction reverses and water starts entering the polders again. As per the MIKE11 simulation, the entire emptying process takes about 24 days. The long duration of the emptying process is because after a certain time the water level of the polders become same as the water level in the Elbe and hence, the water levels in the polder fall along with the river water level. It is to be mentioned here that all gate (c) operations are automatically executed in the MIKE11 model simulations based on the selected decision criteria. As mentioned earlier the storage area under investigation is yet to be constructed and hence, only calibration data for the river is available. Due to lack of calibration/validation data sets for the storage area the simulation results obtained herein are compared with a similar study. In IWK (2003) the peak attenuation effect for several proposed flood storage areas along the Middle Elbe River was simulated considering floods with peak discharges ranging between 4000 m³/s and 5000 m³/s. For the same storage area as investigated in the present study a Fig. 4: MIKE11 simulation results for the proposed peak reduction between 262 m³/s (14 cm) and emergency storage area for the August 2002 flood event 497 m³/s (23 cm) at the gauge Wittenberg was (positive discharge in flow direction from South to simulated. These results are very similar to the North): (a) Simulated discharge and water levels, (b) Gate range of water level reduction obtained in the levels, (c) Gate discharge present study. 48 Comparison of hydrodynamic models of different complexities …

4.3 Comparison of one- and one-/ two- variations of ground elevations. However, a dimensional model results work around is possible in MIKE11 by raising The DEM grid size used for MIKEFLOOD the sill level of the north gate to 73.5 m when is 50 m and the area-elevation curves for the water level in the north polder lowers to MIKE11 are also extracted from 50 m grid 73.5 m during the emptying process. However, DEM. A comparison of the results obtained this would require the use of MIKEFLOOD from MIKE11 and MIKEFLOOD simulation model to ascertain the required water level (73.5 runs show that there is absolutely no difference m in this study) prior to using MIKE11. Such in the water level and discharge reduction in the an approach was not adopted herein as this pa- Elbe River. This is because the discharge per aims at an independent comparison of the through the south gate is the same for both 1D and 1D-2D models to model floods with models. The identical discharge is because it is a emergency storage areas. case of free flow discharge controlled by the MIKEFLOOD results for the polders show upstream water level and the upstream water large tracts of agricultural land, particularly in level for the south gate for both the models is the northern side of the north polder (with same even though the downstream water level depths of water as high as 1.5 to 2 m in some differs. places) remain inundated after the emptying The differences between MIKE11 and process through the north and south gates. Be- MIKEFLOOD results are that for MIKE11 the cause of the topography, this water cannot be water front reaches the connecting gate at the drained using the gravity process through the same instant at which the south gate is opened gates. Hence, some of the water may be drained while for MIKEFLOOD the water front takes using a small gate in a stream on the northern about an hour to reach the connecting gate. boundary (not considered here in the modeling Further, due to the different treatment of the process) and the rest of it has to be pumped out polder filling in the models, the water levels up- or gradually evaporate or seep away. stream and downstream of the connecting gate An additional information which is obtained differ for the two models. As a result there is a from MIKEFLOOD is the water velocities in slight difference in the discharge through the the polders. It is observed that at some places connecting gate for the two models. in the polder near the south gate, the velocity is In the emptying process of the polders there higher than the mean velocity of 1.5 m/s (for is significant difference between MIKE11 and 50 m grid size DEM). This type of information MIKEFLOOD results. For the MIKEFLOOD will be of particular help in studying the erosion model, the emptying process continues till the and sedimentation problems in the polder as water level in the polders lowers down to about well as in the subsequent risk analysis. 73.25 m. The emptying process takes about 4 days with most of the emptying taking place in 4.4 Computation time, storage space the first one and a half days. The emptying requirements and modelling effort process stops after the water level reaches 73.25 A comparison of the computational time re- m because the ground elevations near the north quirements for the two models was carried out. gate are higher than its sill elevation (70.8 m) For this the models were run in a personal and this does not permit further draining of the computer having AMD Athlon(tm) 64 3500+ water to take place. However, for the MIKE11 processor with 2.2 GHz speed and 2GB RAM. model, the emptying process continues till the Also, the models were run for the filling and water level in the polders lowers to the sill ele- emptying processes in the polders as well as vation of the north gate (70.8 m) along with the flow in the Elbe River for the same August river water level. The emptying process takes 2002 flood event. The MIKEFLOOD (with 8 about 24 days. This emptying result of MIKE11 m grid DEM for the polders) model was run is in fact incorrect since practically the draining for shorter durations because of very high process cannot continue below the water level computational time and storage space require- of 73.5 m because of the ground elevation con- ments. The simulation time step interval and re- ditions near the north gate as mentioned above. sult storing frequency for the different runs are Such an error is expected to occur in MIKE11 shown in Table 4. The computational time as because the area-elevation curves which de- well as storage space requirements for the scribe the polders do not take care of the spatial Chapter IV 49 model runs is shown in Table 5. It is observed DEM near its links with the structures of that the computation time as well as the storage MIKE11 to bring about model stability. The space requirements for MIKE11 model is very DEM is cut and levelled near the structures and low while these are very high for the provided with an initial water level. In compari- MIKEFLOOD model. As expected, for son, considerably less effort is required in set- MIKEFLOOD the computation time and stor- ting up the MIKE11model. age space requirements increase drastically when finer resolution DEMs are used. Tab. 5: Computation time and storage space requirement for different model runs Tab. 4: Model run details for the August 2002 flood DEM grid Computation Storage event Model size time (h:min) space MIKE11 2 min 22 MB Model Simulation time Result storing step interval (s) frequency (min) 50 m DEM 3 h 43 min 1.1 GB MIKE11 5 s 5 MIKEFLOOD 25 m DEM 14 h 23 min 3.1 GB MIKE11 – 2 s 2 MIKEFLOOD MIKE21 – 2 s 15 8 m DEM* 12 h 40 min 1.3 GB * The MIKEFLOOD model with 8 m grid DEM is As far as the modelling effort is concerned, simulated only for the polder filling process i.e. from 5th considerable effort is required in setting up of Aug. to 21st Aug., 2002. the MIKEFLOOD model. For MIKEFLOOD, quite a few adjustments had to be made in the

n (river) = 0.0361; n (floodplain) = 0.0475 n (river) = 0.038; n (floodplain) = 0.05 79 n (river) = 0.0399; n (floodplain) = 0.0525

77 South Gate at 182.6 km Water level, m Water 76 Mauken Gauge at 184.4 km

74 174 177 180 183 186 189 192 195 River chainage, km

Fig. 5: Maximum water levels along longitudinal section of Elbe River as obtained from M11 (polders are not considered) for different ‘n’ values for the August 2002 flood event

50 Chapter IV

Considering the fact, that the maximum water level reduction at the Mauken gauging site is 25 4.5 Sensitivity analysis cm (as stated earlier), these water level differences of 12-15 cm due to change of n values 4.5.1 Manning’s n seem to be significant. Figure 5 shows the maximum water levels Figures 6a and b show the results of sensi- along the longitudinal section of Elbe River as tivity analysis of the MIKE11 model to the obtained from MIKE11 (when only the Elbe Manning’s n values when the polders are in- River is modeled and the polders are not concluded. It is observed that when the n values sidered) for different ‘n’ values for the August are decreased by 5%, the water level reduction is only 12.3 cm and discharge reduction is 137 2002 flood event. It is seen that as the n values 3 are decreased the water level decreases and m /s (Figure 6a). While the water level reduction is only 17.0 cm and discharge reduction is vice-versa. When the n values are decreased by 3 5%, the maximum water level difference occurs 218 m /s when the n values are increased by at the upstream end which is 16 cm while the 5% (Figure 6b). This happens because though water level differences at the points of interest the south gate is still opened at the same water i.e. at the south gate is 15 cm and at the Mau- level value of 76.94 m, the river discharge cor- ken gauging site is 13cm. Similarly, when the n responding to this water level is different for values are increased by 5%, the maximum water the two cases due to different n values. As stated earlier, the water level reduction is 25 cm level difference also occurs at the upstream end 3 which is 15 cm while the water level differences and discharge reduction is 310 m /s when the at the points of interest i.e. at the south gate is calibrated values of n are used. Thus, the model 14 cm and at the Mauken gauging site is 12 cm. is quite sensitive to changes in n value.

(a)

(b)

Fig. 6: Sensitivity of MIKE11 model to Manning’s n values when polders are considered: (a) n = 0.0361 for river and 0.0475 for floodplain, (b) n = 0.0399 for river and 0.0525 for floodplain Chapter IV 51

In this study, as mentioned earlier, the Man- cases the polders are filled to their design levels ning’s n values obtained during the calibration i.e., 76.14 m for south polder and 75.35 m for and validation process are 0.038 and 0.05 for the north polder. Though the discharge the main channel and adjacent floodplain, re- through the south and connecting gates are spectively. The corresponding normal values of same for both cases, the gates close a little ear- Manning’s n for the prevailing land-use men- lier for 50 m DEM case than 8 m DEM case. tioned in literature (Chow, 1959) are 0.035 (for This minor difference in the result is due to the natural streams – major rivers) and 0.05 (for slightly different volume-elevation curves de- floodplains – light brush). As the calibrated val- rived from the two DEMs. Because of averag- ues are very close to those mentioned in the lit- ing, the 50 m DEM has a slightly lesser volume erature and the land-use in the study area is for the design water level as compared to the 8 quite uniform, a lower range (± 5%) of Man- m DEM. For the 50 m DEM, the volume of ning’s n is used in the sensitivity analysis. The water corresponding to the design water levels uncertainty associated with the roughness val- are 20.24 Mm3 in the north polder and 20.32 ues in modelling floods has been a subject of Mm3 in the south polder (i.e. a total volume of continuous research (Werner et al., 2005). Hor- 40.56 Mm3). Whereas for the 8 m DEM, the ritt (2005) states the difficulty in specifying the volume of water corresponding to the design hydraulic roughness values in spite of having a water levels are 20.44 Mm3 in the north polder reasonable idea of the land-use. The author fur- and 20.54 Mm3 in the south polder (i.e. a total ther suggests the use of calibration approach to volume of 40.98 Mm3). Thus, the total differ- remove this difficulty. Hence, it is proposed ence in volume of polders for both cases is 0.42 that a more detailed calibration and validation Mm3. But this does not produce significantly procedure be adopted considering a large num- different results. Hence, a 50 m DEM can very ber of flood events in order to reduce the un- well be used to derive the area-elevation curves certainties associated with the Manning’s n val- for the 1D MIKE11 model and yet get accurate ues. However, it is also expected that the results. sensitivity to n values would decrease when The results of sensitivity analysis of more than one polder is used and the conse- MIKEFLOOD model to the use of different quent peak water level reduction in the Elbe DEM resolutions for the polders also show that River is much higher. the water level and discharge reduction in the The results of sensitivity analysis of the Elbe River remain the same. However, when MIKEFLOOD model show that they are in- the south gate is closed after the filling process, sensitive to the variation of n values in the pol- the discharge in the Elbe River for the 8 m ders. This is quite expected because (i) the in- DEM case increases to a lesser extent than that flow to the polder remains the same as it is not of the 25 m DEM case which in turn increases influenced by the polder water level and (ii) n is to a lesser extent than that of the 50 m DEM proportional to the velocity which in bulk char- case. This is because for the design water level, acteristic is low. This finding justifies using only the volume of 8 m grid DEM is slightly higher one roughness value for the polders rather than than that of the 25 m grid DEM which in turn differentiating into several roughness classes. is higher than that of the 50 m grid DEM. As a Similar results are also reported by Werner et al. result, for the 50 m DEM case the south gate (2005). closes ahead of the 25 m DEM case which in turn closes ahead of the 8 m DEM case. The 4.5.2 DEM’s of different resolutions south gate discharge is same in all cases because of same upstream water level. So, even though The results of sensitivity analysis of the downstream water levels differ, the dis- MIKE11 model to area-elevation curves de- charge remains the same as it is a case of free rived from different grid size DEMs show that the water level and discharge reduction in the flow discharge governed by upstream water Elbe River remains unchanged. However, when level. The water front takes about an hour to the south gate is closed after the filling process, reach the connecting gate for all cases. How- the discharge in the Elbe River for the 8 m ever, the discharge through the connecting gate DEM case increases to a lesser extent than that is slightly different for the three grid size DEMs of the 50 m DEM case. For both the DEM because of varying upstream and downstream 52 Comparison of hydrodynamic models of different complexities … water levels for the three cases. The upstream 3.36 m (Figure 7b). For the 8 m DEM, the in- water level for the 8 m case remains lower than undation extent is 1.23 km2 and the maximum for the other two since the 8 m DEM has the water depth is 3.64 m (Figure 7a). For the 50 m same volume of water at a lower elevation DEM, the surface elevations are higher than the compared to that of 25 m and 50 m DEM. 25 m and 8 m DEM. Hence, the maximum wa- Figures 7a to c shows the flood inundation ter depth is the lowest for 50 m DEM and the extent and depth in the south polder for the resulting inundation extent is the highest. Fur- three DEM cases (8 m, 25 m and 50 m) on 17th ther, though the total inundation extent for the August 2002 at 16.30 hours, i.e. 40 min after 8 m and 25 m DEM are same, their spatial the filling process starts through the south gate. variation is different, particularly at the fringes At this instant of time, the volume of water that (Figure 7a and b). Unlike floodplain inundation enters the south polder is the same for all the studies, for polder studies, the analysis of the three cases as the discharge through the south inundation extent and depth for different gate is the same for all cases. For the 50 m DEMs is not of much significance since after DEM, the inundation extent is 1.35 km2 and the initial phase where the water front pro- the maximum water depth is 3.01 m (Figure gresses, the polders begin to fill up and the 7c). For the 25 m DEM, the inundation extent DEM resolution does not play a major role. is 1.23 km2 and the maximum water depth is

(a) (b)

Inundation area = 1.23 km2 Inundation area = 1.23 km2 Maximum water depth = 3.64 m Maximum water depth = 3.36 m

(c)

Fig. 7: Flood inundation extent and depth on 17th August 2002 at 16.30 hours in south polder as obtained from MIKEFLOOD for DEM with grid sizes (a) 8 m (b) 25 m 2 Inundation area = 1.35 km and (c) 50 m Maximum water depth = 3.01 m Chapter IV 53

higher and sometimes a little lower than the 4.5.3 Number of cross-sections used case when all 34 cross-sections are used. Figure 8 shows the maximum water levels Whereas, the water levels for case-II are in gen- along the longitudinal section of the river as ob- eral a little lower than the case when all 34 tained from MIKE11 for the case when all the cross-sections are used. However, for the lower 34 cross-sections are used and for the two dif- reaches of the river, the water levels for both ferent cases of cross-sections used for the Au- cases I and II almost coincide with the water gust 2002 flood event. It is observed that for level for the case when all 34 cross-sections are case I, the water levels are sometimes a little used.

All 34 c/s (Actual) 79 14 c/s removed (Case I) 14 c/s removed (Case II)

77 South Gate at 182.6 km

Water level, Water level, m Mauken Gauge 76 at 184.4 km

74 174 177 180 183 186 189 192 195 River chainage, km

Fig. 8: Maximum water levels along longitudinal section of Elbe River as obtained from M11 for different sets of cross-sections for the August 2002 flood event

The water level differences at the points of very well represented by the remaining 20 interest i.e. south gate and Mauken gauging site cross-sections. Thus, in general it can be seen for the different cases are shown in Table 6. It that the number of cross-sections used in this is seen that the water level differences are not study to model the Elbe water level is reasona- that significant considering that 14 cross- bly sufficient. sections are removed. 4.5.4 Gate opening time and opening/closing duration Table 6: Water level differences (in m) in the Elbe River As mentioned earlier, during the filling proc- due to use of different sets of cross-section data ess of the polder, the south gate is opened at No. of Cross- Water level difference (m) at 15.40 hrs on 17th Aug, 2002 in order to obtain a Case sections river chainage maximum water level reduction of 25 cm in the removed 182.6km 184.4km Elbe River. When the south gate opens 6 h I 14 -0.05 -0.02 ahead of this opening time at 9.40 hrs on 17th II 14 -0.09 -0.08 Aug, 2002 (corresponding to the Elbe water level of 76.71 m at the Mauken gauge instead of These results indicate that for the two cases 76.94 m), the water level reduction decreases to when 14 cross-sections are removed, the shape 14.8 cm (from 25.0 cm) which corresponds to a of the river including its depth and width are discharge reduction of 182 m3/s. Similarly, 54 Comparison of hydrodynamic models of different complexities … when the south gate opens 6 h after the actual 5 CONCLUSIONS opening time at 21.40 hrs on 17th Aug, 2002 (corresponding to the Elbe water level of 77.10 For the August 2002 flood event, the poten- m at the Mauken gauge instead of 76.94 m), the tial polder with the proposed gate dimensions water level reduction decreases to 8.6 cm (25.0 and gate control strategy is capable of reducing cm) which corresponds to a discharge reduction the peak water levels near the Mauken gauging of 93 m3/s. This shows the importance of a site in the Elbe River by about 25 cm while the very good forecast for an effective reduction of corresponding discharge reduction is about 310 water levels in the main river. m3/s. The time of opening of the south gate The results of sensitivity analysis of during the polder filling process is decided us- MIKE11 model to different gate opening a trial and error process so as to maximise ing/closing durations during the polder filling the water level reduction in the Elbe River. The process show that for case-I, there is a sudden water level reduction can be further improved fall in the Elbe River discharge (as compared to through different gate control strategies. This the 30 min duration case) when the south gate aspect as well as the effectiveness of the polders opens. This is because the south gate opens in reducing the water levels in the Elbe River faster and hence, the initial discharge through for floods of different magnitudes and duration the south gate is higher. Also, the south gate is discussed in a separate paper by the same au- closes earlier (than for the 30 min duration thors (Förster et al., 2008). As far as the empty- case). This is because both the south and con- ing of the polders are concerned, there are not necting gates are closed when the respective de- much intricacies involved. The emptying proc- sign water levels are reached in the polders. As ess starts when the discharge in the main river the gates close very fast for case-I, the water falls to a low threshold value. level (and hence, the volume) in both the pol- Both the 1D and coupled 1D-2D model ders after the gates are fully closed are lower as simulations for the potential polder yield the compared to the 30 min case. The final volume same water level and discharge reductions in of water in the north and south polders for the Elbe River. However, due to difference in case-I are 20.10 Mm3 and 20.08 Mm3, respec- treatment of the polders in both the models, tively; while the final volume of water in the the results for the flow processes in the polders north and south polders for the 30 min case are are slightly different. For example, there are dif- 20.32 Mm3 and 20.24 Mm3, respectively. Thus, ferences in the time for the water front to reach as the storage volume in the polders is a little the connecting gate as well as the discharge less for case-I, the discharge and water level at through the connecting gate. Also, the empty- Elbe River 184.4 km rises a little higher than ing process of the polders differs significantly the 30 min case. However, the total discharge for the two models. While the 1D model drains and water level reduction in the Elbe River for the polders completely in 24 days, the 1D-2D case-I is the same as that for the 30 min case, model drains it only partially in 4 days. The 1D- because the total discharge and water level re- 2D model result is practically correct as the ductions are still governed by the threshold dis- polders cannot be drained below a certain water charge and water levels at which the south gate level because of ground elevation conditions opens, and this threshold discharge and water near the gates. The 1D-2D model provides ad- level are the same for both the cases. Similarly, ditional information in terms of the areal extent for case-II, as the gates open and close slowly as well as depth of water in the polders after the the final volume of water in the north and emptying process as well as the water velocities south polders are 20.58 Mm3 and 20.43 Mm3, in the polders. The information on velocities respectively. Thus, as the storage volume in the will be particularly useful in studying the ero- polders is a little more, the discharge and water sion and sedimentation problems and subse- levels at Elbe River 184.4 km rises a little lower quent risk analysis in the polders. The computa- than the 30 min case. However, in this case tional time as well as the storage requirements also, the discharge and water level reductions for the 1D model is very less while this is sig- are the same as that for the 30 min case. nificantly higher for the 1D-2D model and more so when finer resolution DEMs are used. Further, unlike the 1D model, considerable effort is required in setting up and simulating the Chapter IV 55

1D-2D model. In view of all these, it is recom- A different gate opening time for the south mended to use a 1D model for studying the gate causes the water level reduction to de- flooding processes of polders, particularly the crease drastically. This indicates that it is very water level and discharge reductions in the main essential to have a good flood forecast in order river. The computational time requirement sug- to effectively reduce the water levels in the gests that a 1D model may be used in a near main river. The change in gate opening and real time mode as well. However, a 1D-2D ap- closing durations from 5 min to 60 min does proach may be used when the study of flow dy- not have an effect on the water level reductions namics in the polder is of particular interest. in the Elbe River. In this study, the gate open- The 1D model is quite sensitive to changes ing and closing duration of 30 min is selected in the values of Manning’s n for the river and based on information provided by the local wa- its floodplain within the embankments. Thus, ter authorities. there is a need for a rigorous calibration and validation of the model before it is put to use. The 1D-2D model is not very sensitive to ACKNOWLEDGEMENTS change in the Manning’s n values for the polders. This is because the ‘n’ values do not have The research was jointly funded by the Alex- a role to play once the water front reaches the ander-von-Humboldt Foundation Fellowship boundary of the polders and the water level in Program and the Sixth Framework Program of the polders starts rising. the European Commission (FLOODsite pro- A coarse resolution DEM can very well be ject, EC Contract number: GOCE-CT-2004- used to derive the area-elevation relationship 505420). This paper reflects the authors’ views for the polders for use in the 1D model and yet and not those of the European Community. obtain accurate results. The same holds true for Neither the European Community nor any a coupled 1D-2D model wherein a coarse reso- member of the FLOODsite Consortium is li- lution DEM for the polders can be effectively able for any use of the information in this pa- used. This would result in significant reduction per. of the computational time and storage space re- Data were kindly provided by the following quirements. In this study, the use of a 50 m grid authorities: Landesbetrieb für Hochwasser- DEM was found to yield good results. schutz und Wasserwirtschaft Sachsen-Anhalt, The number of cross-sections should be Wasser- und Schifffahrtsamt Dresden and Lan- chosen such that the shape of the river includ- desvermessungsamt Sachsen-Anhalt. The au- ing its depth and width are very well repre- thors are also grateful to the Danish Hydraulic sented by them. In this study, it is seen that the Institute (DHI), Denmark for providing an 34 cross-sections used to model the Elbe water evaluation copy of the MIKE software. levels is quite sufficient.

Chapter V

Simulation of water quality in a flood detention area using models of different spatial discretisation

ABSTRACT: Detention areas are used to lower peak discharges during extreme flood events by temporary storage of excess water. Hence, the risk of dike failures and extensive inundations in adjacent and downstream river reaches is reduced. However, ecological side effects such as a deterioration of water quality during water retention may occur. This is mainly due to the large amount of organic matter in the flood water and the inundation of terrestrial vegetation in the detention area. Decay processes can cause a severe depletion of dissolved oxygen (DO) in the temporary water body. The impact of water retention on the DO dynamics in a planned detention area at the Elbe River (Germany) is studied by means of water quality modeling. Models of different spatial discretisation, a zero-dimensional (0D) and a two-dimensional (2D) approach, were applied to assess their suitability in terms of performance and modeling effort. Both model approaches solely differ in their spatial discretisation, while conversion processes, parameters, and boundary conditions were kept identical. The dynamics of DO simulated by the two models are similar in the initial flooding period but diverge when the system starts to drain. The deviation can be attributed to the different spatial discretisation of the two models, and hence the different approach of determining flow velocities and water depths. The 2D model requires significantly higher efforts for pre- and post-processing and longer computing times. It is therefore not suitable for investigating various flood scenarios and for testing the model's reliability with an extensive sensitivity analysis. However, studying the impact of the spatial variability on the evolution of the state variables necessitates a spatially distributed model approach. For practical applications, it is recommended to firstly set up a fast-running model of reduced spatial discretisation, e.g. a 0D model. Using this tool, the reliability of the simulation results should be checked by analyzing the impact of uncertain parameters of the water quality model with a particular focus on those parameters that are spatially variable and, therefore, assumed to be better represented in a 2D model. The benefit from the application of the more costly 2D model should be assessed, based on the analyses carried out with the 0D model. A 2D model appears to be preferable if the simulated detention area has a complex topography, flow velocities are highly variable in space, and the parameters of the water quality model are well known.

Published as: Kneis D, Förster S and Bronstert A. Simulation of water quality in a flood detention area using models of different spatial discretisation. Ecological Modelling. (in preparation) 58 Simulation of water quality in a flood detention area …

The ecological submodel, covering the state variables and interactions shown in Figure 1 is, 1 INTRODUCTION however, identical. The objective is to point out pros and cons of the two levels of discretisation Detention areas form part of the flood man- and to assess the suitability of either approach agement strategy for many lowland rivers such to predict DO concentrations in a flooded de- as the Elbe River (Germany). They are used to tention area. lower peak discharges in the river during large flood events. Excess water is temporarily stored, hereby reducing the flood hazard at adjacent downstream reaches. The utilisation of flood detention areas has been proved to successfully reduce the peak discharge during the Elbe flood in 2002 (Förster et al., 2005). How- ever, such areas are often used for agriculture and, apart from economic losses due to the inundation of crops, water quality problems may occur. The decay of submerged biomass causes high oxygen consumption rates and may lead to a severe depletion of dissolved oxygen (DO) levels. Large amounts of organic matter con- Fig. 1: State variables of the water quality model. DO tained in the river water during extreme floods represents the concentration of dissolved oxygen, MM is the concentration of 'mobile' organic matter in the water (IKSE, 2004) also contribute to oxygen con- column and PY is the symbol for phytoplankton (all in sumption. When a large detention area at the g m 3 ). IM represents the areal concentration of Elbe River was flooded in 2002, fish extinction 'immobile degradable matter at the reservoir's bottom due to DO depletion was nearly 100% (Böhme and SPY is the concentration of settled phytoplankton (both in units of g m 2 ). DO , MM , and PY are et al., 2005). controlled by external forcings (dashed arrows). Water quality modeling provides a means to Interactions between the state variables, including study the impact of flooding on ecologically negative feedbacks, appear as solid lines. Details on the relevant parameters, such as dissolved oxygen. modeled processes are given in Section 3.2. For example, different hydrological scenarios can be simulated to examine the effect of flooding depth, water residence time, and initial 2 STUDY SITE vegetation cover. To produce meaningful predictions, the water quality model must provide The test site for model comparison is a both a reasonable description of the turnover flood detention area to be built at the Middle Elbe River (Germany; 51°43' N, 12°54' E). The processes (ecological submodel) as well as an 2 appropriate representation of heterogeneity, i.e. area with a total extent of 17 km is separated spatial discretisation. What kind of discretisa- from the River by the main dike. It consists of two linked reservoirs with a total storage tion is appropriate, is determined by the hydro- 3 dynamic situation (geometry and boundary capacity of 40 million m (Figure 2). The conditions) as well as the relevant turnover shallow reservoirs are designed to temporarily processes. In practice, however, the actual store part of the discharge of the Elbe River in choice is often dictated by the cost of setting up the case of flood events with a return period a multi-dimensional and/or high-resolution > 100 years. Filling of the reservoirs and post- simulation model and - more often - by data event drainage are regulated by means of the availability. three control structures shown in Figure 2. In this study, we explicitly investigated the Currently about 90% of the area is under in- impact of model discretisation. For that pur- tensive agricultural use with main crop types pose, we simulated the oxygen dynamics in a being grain crops, corn, canola, and intensive planned flood detention area using two models grassland. The remaining 10% are taken up by a in parallel: a zero-dimensional approach (0D small watercourse and adjacent wetland forests, model) and a two-dimensional one (2D model). most of which are under protection according Chapter V 59 to national and European environmental legis- 3 METHODS AND DATA lation. It is expected that the land will retain its original purpose after designation as a detention area. 3.1 Comparison of the 0D and the 2D approach All the basic features of the 0D and the 2D modeling approach are summarised in Table 1. Since we aimed at studying the impact of the models' spatial discretisation, we kept all features except discretisation identical. Conse- quently, the set of state variables and processes is the same in both the 0D and the 2D approach (Figure 1). The two models simulate the evolution of concentrations in a control volume by solving a common set of ordinary differential equations (Table 2, Eq.s 4-20) for a common set of boundary conditions (Section 3.4).

Fig. 2: Map of the flood detention area at the Middle Elbe River. The arrows used for labeling the locations of weirs indicate the normal flow direction.

Tab. 1: Comparison of basic features of the 0D and 2D model. 0D model 2D model Hydrodynamics and transport Spatial discretisation 2 stirred tank reactors 50 x 50 m grid model Flow model Continuity eqn. 2D St. Venant eqn. Geometry input Storage functions DEM as grid Transport model Mass balance 2D Advect.-Disp. eqn. Time step variable, 2 min variable, 2 sec a

a Boundary conditions In- & Outflow rates; see Section 3.4 Water quality model Components & processes User-defined; see Section 3.2 Numerical solver LSODA b 5th order Runge-Kutta

b Time step variable, 2 min variable, 2 min Boundary conditions In-river concentrations, meteorology; see Section 3.4 Computational aspects Software environment R-Script c MIKE, Ecolab d

c d Computation time e ~ 2.5 min ~ 5 hours

e Size of output files f ~ 0.8 MB ~1.7 GB f a Required to ensure Courant numbers < 1 b From R-package ‘odesolve’; Originally by Hindmarsh (1983) and Petzold (1983) c Implemented by D. Kneis using R version 2.6 (R Development Core Team, 2007) d Developed by the Danish Hydraulic Institute (DHI, 2003) e Includes simulation and post-processing on a PC with 2.2 GHz CPU and 2 GB RAM f Uncompressed ASCII text; storage interval 15 min

The difference between the two model is assumes vertical mixing of the water column solely due to a very different discretisation of (Figure 3, left). In the 2D model, each grid cell the model domain as illustrated by Figure 3. with the extent   50== myx has its individual The 2D model accounts for lateral heterogenei- depth D as determined from the water surface ties in hydrodynamic variables and concentra- elevation h and the digital elevation model tions. It solves the advection-dispersion equa- (DEM). Turnover computations are carried out tion to simulate lateral transport but still for each cell. 60 Simulation of water quality in a flood detention area …

Fig. 3: Discretisation of a modeled reservoir in the vertically averaged 2D model (left) and the 0D stirred tank model (right). Symbol h represents the water surface elevation and D is the depth of a grid cell (2D model) or the average depth of the stirred tank (0D model). For this study we used x y 50== m.

In contrast, in the 0D model, each of the the 2D water quality model. The computation two reservoirs shown in Figure 2 is considered of flow rates and depths is based on a finite as a stirred tank reactor (short: STR; Chapra, difference approximation of the Saint Venant 1997). A STR is assumed to be both laterally Equations. and vertically mixed (Figure 3, right). There- fore, it is characterised by homogeneity with respect to all state variables, e.g. concentrations. As opposed to the 2D model, the 0- dimensional stirred tank approach is built on aggregated information on the water body's geometry in the form of storage functions (Fig- ure 4). The basic relation is the one between the water surface elevation h and the surface area A hA )( . is easily computed from a DEM. The relation for the storage volume hV )( as well as Fig. 4: Storage functions representing the upper for the depth hD )( derive from hA )( according reservoir's geometry in the 0-dimensional stirred tank to Eq.s 1 and 2. Here, in contrast to the 2D ap- approach. hD )( proach, is an average value that is assumed The computed inflow rates to the upper to be representative for the STR, i.e. the entire reservoir, exchange flows at the connecting reservoir. weir as well as outflow rates at the lower weir

=hz (see Figure 5 in Section 3.4) served as boundary )(=)( dzzAhV z 0= (1) conditions for the water quality simulations in both the 0D and the 2D model. However, only hV )( the latter approach uses the computation results hD =)( (2) hA )( for each grid cell (velocity components, depth) in the simulation of lateral transport as well as Flooding and emptying of the proposed turnover. detention area were simulated with the MIKEFLOOD hydrodynamic modeling 3.2 Water quality processes package (DHI, 2004). The model integrates a one-dimensional representation of the Elbe The control of the state variables (Figure 1) River (~ 20 km reach) and a two-dimensional by conversion processes and boundary model of the detention area itself. The 2D part conditions is best illustrated by a Peterson of the model uses a rectangular grid with a matrix (Reichert et al., 2001) as shown in resolution of 50 x 50 m for compatibility with Table 2. The actual dynamics of the state Chapter V 61 variables are determined by the process rates r Table 2). The rate expressions are presented in according to Eq. 3, where Yi is the i-th state the subsequent paragraphs (Eq.s 4-20). variable, n is the total number of processes Yd n considered in the model and the vector of i = , rq kki (3) q td k 1= stoichiometry factors 1=, nki represents the i-th column of the stoichiometry matrix (see

Tab. 2: Process matrix of the water quality model showing the influence of processes on the simulated state variables. The columns 4-10 represent the stoichiometry matrix. It contains the factors q ,ki appea ring in Eq. 3. If q ,ki 0> , the k-th process is responsible for an increase in the value of the i-th state variable whereas q ,ki 0< indicates a decrease according to Eq. 3.

For clarity, a period is displayed in those positions where a state variable is not affected by a process, i.e. q ,ki 0= . The rate expressions for all processes are presented separately in Eq.s 4-20. See Tables 3 , 4, and 5 for the definition of all symbols. # Process Rate M MIM P Y SP Y D O TW V expression ( g m-3) ( g m-2) ( g m-3) ( g m-2) ( g m-3) (°C) ( m3) PY  PY DO  DO TW  TW 1 Inflow Eq. 4MM x  MM .x .x x 1 V V V 2 Outflow Eq. 5V ...... -1

3 Decay of MM Eq. 6-1 . . . f MM . .

4 Decay of IM Eq. 7 . -1 . . IM /Df . .

5 Decay of SPY Eq. 8 . . .-1 PY /Df . .

6 Growth of PY Eq. 9 . . 1 .f PY . .

7 Respiration of PY Eq. 12 . . -1 . f PY . . 8 Settling of PY Eq. 13 . .-1D . . . 9a Aeration 1 Eq. 15 . . . .1 . . 9b Aeration 2 Eq. 17 . . . .1 . . 10 Heatflux Eq. 20 . . . . .1 .

Process rates 1 and 2: In- and Outflow (g m 3 ) is the half-saturation constant in the r The process rate 1 accounting for the inflow Monod term accounting for in hi bition of the of water to the reservoir (OD model) or a con- aerobic process at low oxygen levels. The unit r 3 1 trol vo lume (2D model) is nothing but the flow of the process rate 3 is g m s . Q rate in (Eq. 4). Likewise, the process of out- r (TW 20) DO flow 2 is described by the corresponding flow r = MM tk  3 MM MM  hDO (6) Q DO rate out (Eq. 5). The unit of the two rates is 3 1 m s . Process rate 4: Degradation of immobile organic matter r1 = Qin (4) r The rate 4 as defined by Eq. 7 controls the degradation of immobile organic matter that is = Qr 2 out (5) attached to the reservoir's bottom. In Eq. 7, IM is the areal concentration of the degradable ma- Process rate 3: Degradation of mobile 2 D organic matt e r terial (g m ), D (m) is the water depth and min is a threshold value of the depth at which die- r3 The rate describing the velocity of the off and decay of the vegetation sets in. The MM degradation of is given by Eq. 6. Here, k t MM is the concentration of mobile org a nic meaning of the constants IM and IM is equiva- 3 1 lent to the corresponding parameters in Eq. 6. kMM matter in the water column (g m ), (s ) is They were introduced to allow for different de- t the rate of aerobic decay, and MM (-) controls gradability of IM as compared to MM . The re- the process' dependence on water temperature TW DO hDO TW DO maining symbols , , and are the same (°C) through an Arrhenius term. 2 r3 IM 3 h as in (Eq. 6). Since is defined in g m the (g m ) is the oxygen concentration and DO r 2 1 unit of rate 4 is g m s . 62 Simulation of water quality in a flood detention area …

e 1 cient back (m ), and the chlorophyll-a content  TW 20)( DO f 1  tkIM IMIM  >for DD min of the phytoplankton chla (g g ). The latter two r4 =   hDO DO (7)  0 for  DD parameters are used for computing the total ex-  min e 1 tinction coefficient tot (m ) according to Eq. Process rate 5: Degradation of settled 11 (adapted from Ambrose et al., 2001). Light phytoplankton adaption is not taken into account, i.e. f = const The degradation of settled phytoplankton chla . r SPY ( 5 , Eq. 8) is computed in a similar way as the decay of IM (see Eq. 7) and the rate unit is 2.718   I   I  ilim = exp srf  (eexp  D)  exp srf  2 1 IM SPY  De   I tot   I  (10) g m s as well, because, like , is an tot   opt   opt  2 areal concentration (g m ). To allow for differ- (2/3) kSPY ee  8.8= PY  f  (5.4 PY  f ) ent degradability, a separate rate constant ( ) tot back chla chla (11) t and temperature correction factor ( SPY ) were introduced. Process rate 7: Respiration of phytoplankton DO This process accounts for the loss of phyto- = tkSPYr TW 20)(  5 SPY SPY  hDO (8) plankton biomass by respiration and mineralisa- DO r tion of dead algae. The rate 7 with the unit Process rate 6: Growth of phytoplankton g m 3 s 1 is computed according to Eq. 12. The Phytoplankton growth is considered in the k 1 t rate constant resp (s ) and the parameter resp (-) model as is compensates for oxygen consump- are equivalent to the corresponding constants tion by aerobic decay at daytime. The growth in Eq.s 6-8 and the other symbols were ex- r rate 6 is computed as a function of the current plained in conjunction with Eq. 6. 3 phytoplankton concentration PY (g m ) using 1 kgrow DO a potential growth rate (s ) which defines =  k tPYr (TW 20)  7 resp resp DO  h (12) the rate of cell division under optimum condi- DO tions (Eq. 9). The model accounts for the effects of light limitation by the function ilim (see Process rate 8: Settling of phytoplankton Eq. 10) representing the depth-integrated Steele By the process of settling, phytoplankton equation (Ambrose et al., 2001). The impact of ( PY ) is transferred from the water column to r temperature is described by an Arrhenius term the reservoir's bottom. The rate of transfer 8 t (dimensionless parameter grow ) similarly to has the unit g m 3 s 1 . It is estimated by Eq. 13 Eq.s 6-8. For this study, we assume that phyto- u 1 using an effective settling velocity sett (m s ) plankton growth is not limited by nutrients be- and the water depth D (m). cause the concentrations of nitrogen and phos- phorus in the river are high during flood and usett 8 = PYr  (13) additional nutrients are remobilised from the D inundated soil and the mineralisation of vegetation. Process rate 9: Oxygen flux between water column and atmosphere TW 20)( 6 = grow tkPYr grow  optsrf back chla PYfeDIIilim ),,,,,( (9) The model computes the rate of oxygen exchange through the water surface as a linear Arguments to the function ilim are the inten- function of the DO saturation deficit, i.e. the sity of photosynthetic active radiation just be- difference between the actual DO concentra- I 2 low the water surface srf (W m , boundary tion and the temperature-dependent saturation dosat(TW ) 3 I level (g m , Eq. 14). condition) as well as the optimum intensity opt at which gross growth reaches a maximum. 2 Further arguments to ilim are the cu rrent phyto- dosat 14.652=  0.41022 TW  0.007991 TW (14)  0.000077774TW 3 PY 3 plankton concentration (g m ), the water depth D (m), a background extinction coeffi- Chapter V 63

Two separate approaches are used for esti- 3.3 Computational procedures mating the transfer coefficients (proportionality The system of ordinary differential equations r 3 1 factors). In the process rate 9a (g m s , to be solved by the 0D model consists of 2 n k Eq. 15) the transfer coefficient aer,wind is com- equations in the form of Eq. 3 where n is the puted as a function of wind speed WSP (m s 1 ) number of state variables in a stirred tank (see and water depth D (m) using Eq. 16 that has Table 2). There are 2 n equations because the been found by Banks and Herrera (1977). model is designed to also handle alternating flow directions at the connecting weir shown in )(=  WSPDkDOTWdosatr ),( 9a ,windaer (15) Figure 2. Hence, upper and lower reservoir are simulated simultaneously. Numerical integra- 0.728 WSP 0.317 WSP 0.0372  WSP 2 tion is performed by the stiffly stable LSODA k = (16) ,windaer 86400  D method originally developed by Hindmarsh (1983) and Petzold (1983). r 3 1 In the process rate 9b (g m s , Eq. 17), the In the 2D model, the concentrations are k 1 transfer coefficient aer, flow (s ) is estimated from computed for each grid cell at all time steps by the flow velocity U (m s 1 ) and the flow depth (I) calculating the concentration gradients be- D (m) according to the approach of O'Connor tween the previous and the actual time step ac- & Dobbins (McCutcheon, 1989) displayed as cording to the advection-dispersion conditions, Eq. 18. (II) calculating the concentration gradients for the actual time step according to the water quality process rates, and (III) calculating the result- 9b )(=  , flowaer UDkDOTWdosatr ),( (17) ing concentrations by numerical integration of

1/2 the time step gradients from both the advec- 3.93U k = , flowaer 3/2 (18) tion-dispersion and the water quality differential 86400  D equations (DHI, 2003). The integration method The flow velocity U appearing in Eq. 18 is used is a 5th order Runge-Kutta scheme. variable in space and time. In the 2D model, values of U are available for each grid cell from 3.4 Boundary conditions, parameters, and the hydrodynamic simulation. However, in the initial values 0D model, a rough estimate of U can be derived only. For that purpose, we used Eq. 19 To examine the specific impact of spatial where the denominator represents the wet discretisation, we ran the 0D and the 2D model cross-section of a circular reservoir with depth with exactly the same boundary conditions such 2 as flow rates, external loads, and meteorological D (m) and water surface area A (m ). conditions. A summary of the common driving

QQmax ),( forces is given in Table 3. U = outin (19) As a hydrological boundary condition for  AD /2  this study we adopted the hydrograph of the Process rate 10: Change in water tem- Elbe River from the 2002 event (return period perature ~ 200 years). The gates were assumed to be For this study, the water temperature TW operated in a way that minimises the peak flow was simulated by a simple equilibrium approach in the river while using the detention area's full r storage capacity. Details on hydrodynamic (Eq. 20). The process rate 10 with the unit °C simulations carried out for this and other flood 1 twequi s is controlled by the parameter , repre- events can be found in Chatterjee et al. (2008) senting the ultimate water temperature under and Förster et al. (2008). The relevant flow given meteorological conditions. The rate con- rates at the three weirs (Figure 2) are presented k stant heat determines how fast TW approaches in Figure 5. tw equi . Symbol D (m) represents the water depth.

TWtw = kr  equi (20) 10 heat D

64 Simulation of water quality in a flood detention area …

Tab. 3: Boundary conditions of the water quality models. Symbol Units Description 3 -1 Qin m s Inflow into the reservoir from the river or the adjacent reservoir (0D model) or inflow of a single grid cell (2D model).

3 -1 O out m s Outflow from the reservoir (0D model) or outflow of a single grid cell (2D model).

 1 -3 a MM x g C m Concentration of degradable organic matter (MM ) in the inflow.

a -3 b PY x g C m Concentration of phytoplankton (PY ) in the inflow. -3 DO x g DO m Concentration of DO in the inflow.

TW x °C Temperature of inflow. -2 c I srf W m Photosynthetically active radiation just below the water surface.

 2 c WSP m s-1 Wind speed at the nearest meteorological station.  1 a For inflow from the river, MM x was estimated from observed concentrations of total organic carbon in the Elbe River. f b For inflow from the river, PYx was derived from observed chlorophyll-a in the Elbe River using the value of chla from Table 4. c was computed from shortwave radiation at the nearest meteorological station assuming that 45% of the energy is I srf available for photosynthesis (Ambrose et al., 2001; Romero et al., 2004). An approximate albedo a was computed as a Nsin   /2)/3652(0.020.08= where N is the day of the year (Antenucci and Imerito, 2002).

therefore, observation data for model calibration do not exist. Since this study aims at comparing alternative models rather than at making precise predictions, the lack of calibration data is less severe. However, using the values from Tables 4 and 5, the obtained dynamics and the minimum DO concentration are in good agreement with observations from

Fig. 5: Flow rates at the three gates shown in Fig. 2. other detention areas that were flooded during Positive values indicate flows in the following direction: the same event, i.e. the Elbe flood in 2002 River  upper reservoir  lower reservoir  river. (Knösche, 2003).

The loads of mobile components entering the upper reservoir were computed from the 3.5 Analysis of uncertainty inflow hydrograph at the upper weir (Figure 2) Overall model uncertainty results from the and observed concentrations in the Elbe River combined effect of uncertain input data, model for the respective time period with daily parameters, and structural deficits (Kneis, 2007; resolution (Lindenschmidt et al., 2008). Time Radwan et al., 2004). To get more insight in the series of wind speed and solar radiation were reliability of the simulation results, the 0D adopted from the nearby meteorological station model was integrated in a Monte-Carlo Bethau. environment. We simultaneously varied those In addition to common boundary parameters of the water quality model which we conditions, we used an identical set of considered most uncertain, such as the various parameters and initial conditions in the two rate constants controlling the decay of organic models (Tables 4 and 5). One should note that matter. 250 parameter sets were created with most of the parameter values were taken from the latin-hypercube method using the ranges the literature. This was necessary, because the given in Table 6 and assuming a uniform studied flood detention area is not yet built and, distribution for each varied item. Based on the Chapter V 65 output of all model runs, we calculated different empirical formulas describing the quantiles of the predicted concentrations for dependence of re-aeration on wind speed and every time step. water depth. One of the tested relations is Eq. In addition to the Monte-Carlo simulation 16 and the remaining seven approaches were described above, we run the 0D model with the selected from a list provided by Bowie et al. standard parameters from Table 4 but eight (1985).

Tab. 4: Parameters of the water quality models and standard values used in the comparison of the 0D and 2D approach. Symbol Value Units Description

b -1 f MM 2.76 g g Oxygen-to-carbon ratio for the decay of MM . c -1 f IM 2.8 g g Oxygen-to-carbon ratio for the decay of IM . b -1 f PY 2.76 g g Oxygen-to-carbon ratio for the decay of PY . b -1 f chla 0.028 g g Chlorophyll-a to carbon ratio in phytoplankton. d -1 k MM 0.032 d Rate constant for decay of MM at 20°C. e -1 k IM 0.018 d Rate constant for decay of IM at 20°C. f -1 k SPY 0.032 d Rate constant for decay of SPY at 20°C. g -1 k resp 0.125 d Rate constant of PY respiration at 20°C. a,b -1 k grow 1.8 d Maximum phytoplankton growth rate at 20°C. g -1 u sett 0.1 m s Effective settling velocity of phytoplankton. a -3 h DO 0.5 g m Half-saturation concentration of DO for aerobic degradation of organic matter.

D min 0.03 m Water depth at which vegetation die-off starts. a t MM 1.045 - Temperature coefficient for decay of MM . a t IM 1.045 - Temperature coefficient for decay of IM . a t SPY 1.045 - Temperature coefficient for decay of SPY . g t resp 1.068 - Temperature coefficient for PY respiration. g t grow 1.045 - Temperature coefficient for PY growth. h -2 I opt 145 W m Optimum light intensity for PY growth. i -1 e back 2 m Background light extinction coefficient. j tw equi 20 °C Equilibrium temperature. j -1 k heat 0.2 s Heat transfer rate. a From Bowie et al. (1985) b From Lindenschmidt (2006) c Computed from C/N ratio of biomass. d Estimated from BOD7 and TOC data of the Elbe River making use of fMM . e Derived from experiments undertaken by Peukert (1970). f Set to kMM . g Values rec om mended by Ambrose et al. (2001). h Average of values presented by Bowie et al. (1985); 1 W m 2 = 2.065 Ly d 1 . i Calculated from Secchi depth Ds (m) and Chlorophyll-a data for the Elbe River using Eq. 11 and the approximation

etot 1.7/= Ds . j For this study, the temperature model was practically turned of by setting twequi equal to the water temperature of the river and chosing a large value for k . heat

66 Simulation of water quality in a flood detention area …

Tab. 5: State variables of the water quality models and initial values. Symbol Units Description Initial value M M g C m-3 Concentration of, mainly dissolved, degradable organic matter (as carbon) in the 0 a water column. IM g C m-2 Areal concentration of degradable plant matter at the bottom. 175 b PY g C m-3 Concentration of total phytoplankton. 0 a SPY g C m-2 Areal concentration of settled phytoplankton. 0 DO g m-3 Dissolved oxygen concentration. f (TW) c TW °C Water temperature. TW of river a V m3 Storage volume of the reservoir (OD model) or grid cell (2D model). 1 a A m2 Water surface area of the reservoir (0D) or grid cell (2D). f (V) d D m Water depth; Reservoir-average in the 0D model. d f (V) a The initial value has no effect on simulation results because the reservoir's initial volume is negligible compared to its storage volume after filling. A non-zero initial volume is required because water depth and storage volume appear in the denominator of many expressions (see e.g. Table 2). b Calculated from a plant matter of 600 g DW m-2 with a carbon content of 0.46 g C (g DW)-1 (MLUR, 2007). About 35% of the material was considered as non-degradable within the relevant time span (Peukert, 1970). c The DO saturation level at the inital value of TW was used. d In the 0D model, the values are computed from V based on the reservoir's geometry.

Tab. 6: Parameter ranges considered in the Monte-Carlo simulation carried out with the 0D model. Item(s) Range

Rate constants k mm , k im , k spy , k resp Values from Table 4 ± 50 %

Phytoplankton growth rate k grow Value from Table 4 ± 25 %

Optimum light intensity I opt Value from Table 4 ± 25 %

Phytoplankton settling velocity u sett Value from Table 4 ± 50 %

Chlorophyll-carbon ratio f chla Value from Table 4 ± 25 % -1 Background extinction coefficient e back 1.5-2.5 m

Equilibrium water temperature tw equi 20-25 °C Initial value of IM Value from Table 5 ± 25%

Figure 6 illustrates the change in the upper reservoir's flooding depth over the simulation 4 RESULTS period as a result of the in- and outflow rates (recall Figure 5). In general, the reservoirs average water depth as output by the 0D model 4.1 Simulation results of the two models is close to the median depth computed by the 2D approach. However, due to the reservoir's Figures 6-9 depict the evolution of state natural topography, the actual depth varies variables and process rates as simulated by the between 1.5 and 3.4 m at the time of maximum two models. We only present figures for the storage (10- and 90-percentile). upper reservoir shown in Figure 2 because the Using the same plot setup as in Figure 6, results for the lower one are very similar. To Figure 7 illustrates the simulated concentration illustrate the spatial variability in the 2D model, of dissolved oxygen in the upper reservoir. the median is plotted as well as the 10- and 90- Starting at an equilibrium value of ~ 9 g m 3 percentile of all cells that are wet at a time. before flooding, the inflow of undersaturated Chapter V 67 river water on August 17 causes a first The simulated dynamics of dissolved oxygen significant drop to a DO level of about (Figure 7) is also reflected by the process rates. 6.5 g m 3 . After 4 days of declining For example, Figure 9 illustrates how the rate concentrations, the DO concentration has of degradation of organic matter at the almost reached its minimum at 1 g m 3 on reservoir's bottom develops. Due to the Monod August 21. This minimum occurs shortly after term appearing in Eq. 7, this process rate is the time of maximum storage. Over the directly affected by the concentration of DO. subsequent days, the oxygen concentration Both the computed dynamics as well as the slowly rises and a pronounced diurnal variation deviation between the results of the two models develops. show similarities with Figure 7.

Fig. 6: Water depth in the upper reservoir as simulated by Fig. 7: Concentration of dissolved oxygen in the upper the 0D stirred tank approach and the 2D model (spatial reservoir as simulated by the 0D stirred tank approach median, 10- and 90-percentile). The abscissa's left limit and the 2D model (spatial median, 10- and 90- corresponds to the time when the reservoir becomes percentile). flooded. The triangle marks the time of maximum storage when emptying of the reservoir begins.

During the period of falling DO levels, the results of the 0D model are close to the output of the 2 D simulation. As indicated by the narrow range of the 10- and 90-percentile in Figure 7, the concentration does not show significant spatial variability. At the time of maximum storage, the output of the two models starts to diverge, with higher DO levels being predicted by the 0D approach. Fig. 8: Concentration of phytoplankton carbon in the As can be seen in Figure 8, the predicted upper reservoir computed by the 0D and the 2D model. concentration of phytoplankton continuously rises throughout the simulation period. Du ring the simulated event, the corresponding concentration in the Elbe River amounts to  0.2 g m 3 . In the detention area, the value rises up to 5 g m 3 in the 2D model and 8 g m 3 in the 0D model. According to the stoichiometry factor fchla from Table 4 , this is equivalent to chlorophyll levels of 140 and 244 µg L 1 , respectively. Beginning at the time of maximum storage, the predictions of the two approaches diverge. The reservoir-average phytoplankton Fig. 9: Rates of the degradation of immobile organic level as output by the 0D model becomes matter (Eq. 7) in the upper reservoir as simulated by the similar to the spatial 90-percentile calculated 0D and the 2D model. from the 2D simulation results. 68 Simulation of water quality in a flood detention area …

The 2D model allows us to directly view the below 1.5 g m 3 in both the upper and lower computed spatial heterogeneity of oxygen reservoir. When the reservoirs begin to run dry, concentrations at selected time steps more or less fragmented water bodies of (Figure 10). During the filling period, the different depths develop as indicated by the predicted variability in concentrations is low. At white colors in the plots for the two final days. the time of maximum storage on August 21, the At this stage, we observe increased spatial simulated DO concentrations are generally gradients in the predicted DO levels.

Fig. 10: Dissolved oxygen concentrations in the upper reservoir for the first 6 days of flooding as computed by the 2D model (values at midnight).

The span of simulated concentrations is surprisingly high with differences of up to 4.2 Uncertainty of predictions 5 g m 3 at a time step. Depending on the re- Figure 11 illustrates the range of dissolved aeration formula, minimum DO levels between oxygen concentrations obtained in the Monte- 0.3 and 3 g m 3 are predicted. With our Carlo simulation (Section 3.5). The varied standard approach underlying all other parameters listed in Table 6 control various simulation results presented in this paper processes affecting DO consumption, (Eq. 16, solid graph in Figure 12), intermediate production, and even solubility. While the range results are obtained. between the 10- and 90-percentiles is rather narrow in the period of rising water levels, the span of predictions grows substantially as the reservoir drains and multiple water bodies develop (recall Figure 10). It is worth noting that, in the final period of the simulation, the absolute values but also the range of DO concentrations become particularly large during daytime. The minimum DO concentrations observed at nighttime, however, increase much slower and the spread of the results remains much narrower. Daily minimum values lie Fig. 11: Quantiles of simulated DO concentrations in the between 0.5 and 4 g m 3 DO (10- and 90- upper reservoir from 250 simulations with randomly percentile). modified parameters within the ranges listed in Table 6. The oxygen dynamics as computed with eight different formulas for wind-dependent re- aeration (see Section 3.5) is shown in Figure 12. Chapter V 69

variable re-aeration rates in the 2D model (Figure 13). The fact that the rate of flow- induced aeration in the 0D model exceeds the median of the 2D results during the drainage period (Figure 13, bottom) coincides with the higher DO levels predicted by the 0D approach around August 22 (Figure 7). The observed differences in the flow-induced re-aeration rates are not unexpected because of the necessarily simple approach for estimating U in the 0D

Fig. 12: Dynamics of dissolved oxygen simulated with model (Eq. 19). the 0D model using 8 different empirical formulas for wind-dependent re-aeration. The result obtained with the approach of Banks and Herrera (1977) (Eq. 16) is plotted as solid line. Dashed lines represent the output for the 7 tested alternative formulas listed in Bowie et al. (1985).

5 DISCUSSION AND CONCLUSIONS

5.1 The impact of spatial discretisation

Dissolved oxygen does not only play a central role in the turnover of organic matter (Figure 1), but its concentration also controls the abundance of higher aquatic species such as fish. Hence, we are particularly interested in making good predictions of DO and, consequently, we focus the following discussion on that state variable. According to Figure 7, the DO concentrations simulated by the two models are very similar in the initial period. However, Fig. 13: Rate of wind-induced re-aeration (Eq. 15, top during the period of drainage starting on figure) and flow-induced re-aeration rate (Eq. 17, bottom August 21, the output of the models deviates figure) as computed by the 0D stirred tank approach and more and more. The DO levels in the 0D the 2D model. model permanently exceed the corresponding 90 percentiles computed from the 2D model's Another very important but more indirect results. link between water depth D and the DO To explain the observed differences, we concentration is due to the appearan ce of D in have to consider all processes being directly or the light limitation term controlling indirectly affected by the model's spatial phytoplankton growth (Eq. 10). Given a total extinction coefficient e and a below-surface discretisation. Thus, we have to look at those tot process rates (Eq.s 4-20) and stoichiometry light intensity I srf , there is a depth Dopt , where factors (Table 2) that depend on water depth D ilim has a maximum and so has the growth rate. and/or flow velocity U . These two variables are As long as > DD opt , phytoplankton growth is reservoir-averages in the 0D model, while each limited by light availability. But, according to grid cell in the 2D model has its individual Eq. 10, phytoplankton growth may also be value. light-inhibited if D < D because of increased For example, both U and/or D appear in opt the expressions controlling re-aeration (Eq.s 16 photo-respiration and possibly pigment and 18). Consequently, we observe spatially destruction at too high radiation intensities. In 70 Simulation of water quality in a flood detention area …

our models, we have the situation > DD opt for In the subsequent days of the simulation, the the largest part of the simulation period, i.e. difference between the prediction of the 0D growth is mostly light-limited. model and the spatial median of the 2D results As shown in Figure 6, the reservoir's average is in the order of 1-1.5 g m 3 (Figure 7). At the depth in the 0D model is always close to the same time, the range of concentrations covered median depth computed from the 2D results. by 50% of the Monte-Carlo simulation results is Nevertheless, because of the spatially variable as wide as 2 g m 3 and the span between the 10- depth in the 2D model and the highly non- and 90-percentile amounts to ~ 3 g m 3 linear character of Eq. 10, the average (Figure 11). Thus, the uncertainty in predictions phytoplankton growth rates computed by the due t o insufficient k nowledge of t he model two models diverge considerably, leading to parameters is, again, larger than the error we different phytoplankton levels (Figure 8) and make by neglecting spatial variability, i.e. by different rates of biogenic DO production using the 0D instead of the 2D model. (second part of the simulation period in The uncertainty associated with the Figure 7). simulated DO concentrations becomes even Finally, when interpreting simulation results more obvious when we look at Figure 12. for the drainage period, we must take i nto Apparently, although empirical re-aeration account the evolution of multiple water bodies formulas are part of many water quality models, in the 2D model (Figure 10). Since these wet their use must be considered as a major source patches are only weakly connected or even fully of uncertainty. Depending on the selected isolated, the leveling effect of mixing becomes formula for wind-dependent re-aeration, more and more negligible. In the 0D stirred minimum DO concentrations between about tank model, however, the assumption of 0.5 and 4.5 g m 3 are obtained. This range is complete and permanent mixing is inherent. larger than the greatest spatial difference computed by the 2D model for any time step. 5.2 Conclusions from the uncertainty analysis 5.3 Implications and recommendations If we want to assess the suitability of either model for practical applications, comparing the Our study shows that the parallel use of a spread of results due to the model's 0D and a 2D model is not just of scientific discretisation (Figure 7) to the spread of results interest but is also beneficial for practical due to uncertain parameters and mechanisms applications. (Figures 11 and 12) is particularly interesting. The primary advantage of the zero- Because of the ecological relevance, it is dimensional approach is the short computing advisable to focus on the state variable DO and time as well as the minimum effort for post- the times when the concentration is lowest. processing of the model's output. This is a The minimum DO concentrations at the end precondition when various flood scenarios with of the filling period simulated by the two varying boundary conditions need to be models are almost identical (Figure 7). The examined. Furthermore, only the fast 0D model prediction of the 0D model coincides with the allows extensive analyses of sensitivity/ 2D model's median and the difference between uncertainty to be carried out. We have shown the spatial 10- and 90-percentile is  0.5 g m 3 that, for assessing the reliability of simulation only. However, we can conclude from results, such analyses are absolutely essential. Figure 11 that, due to uncertain rate constants The advantage of the two-dimensional and further parameters of the model, we cannot model is its ability to simulate the heterogeneity actually predict the minimum of DO that with respect to the state variables, i.e. accurate. For example, the interquartile range of concentrations. This is desired if the water body the 250 concentration time series obtained in has a complex topography with significant the Monte-Carlo simulation is in the order of variations in depth, if lateral mixing is 2 g m 3 at the time when the lowest DO levels ineffective, or when there is a pronounced occur. spatial variation in flow velocities. The advantage of a multi-dimensional simulation is, however, gained by much extra effort for the Chapter V 71 preparation, storage, and post-processing of distributed (e.g. 2D) model than in the 0D data as well as for running the model (recall approach. (We skipped this step as we aimed at Table 1). The long computing times resulting testing the 2D model anyhow.) from the spatially explicit solution and the 4. Build a distributed model if the spread of associated restrictions on the length of a simulation results obtained in step 3 is simulation time step may severely limit the significant when compared to the spread of model's overall applicability. results which is due to uncertain parameters of In Section 5.2 we have shown that the water quality model (step 2). knowledge about the consequences of uncertain 5. For detention areas with deep reservoirs model parameters may be quantitatively more the assumption of vertical mixing may be relevant than a proper representation of spatial inappropriate. Based on the experience gained heterogeneity. In other words, the exclusive in this study, we recommend testing a 1D- application of a slow-running, highly discretised vertical model before building a 3-dimensional model, which cannot practically be run in a one. Monte-Carlo environment, is only reasonable, if the parameter values are very well known. In applied water quality modeling this is a rare ACKNOWLEDGMENTS case. For those who aim at modeling water quality The research was funded by the Sixth in a similar context, the following Framework Program of the European recommendations may be helpful: Commission (FLOODsite project, EC Contract number: GOCE-CT-2004-505420). This paper 1. Build a simple zero-dimensional model reflects the authors' views and not those of the first. European Community. Neither the European 2. Use this lightweight tool to check the Community nor any member of the sensitivity of results against uncertain FLOODsite Consortium is liable for any use of parameters and empirical formulas being part the information in this paper. of the water quality model. Check whether the The authors are grateful to K.-E. range of 'likely' results matches the required Lindenschmidt and M. Baborowski for accuracy in the target application, e.g. by scientific support and provision of water quality running a Monte-Carlo simulation. data. Topography and discharge data were 3. Using the 0D model, examine the kindly provided by the federal and state water sensitivity of those parameters that are spatially authorities LHW Sachsen-Anhalt and WSA variable such as flow velocity and depth and Dresden. which are, therefore, better represented in a

Chapter VI

Assessing flood risk for a rural detention area

ABSTRACT: Flood detention areas serve the primary purpose of controlled water storage during large flood events in order to decrease the flood risk downstream along the river. These areas are often used for agricultural production. While various damage estimation methods exist for urban areas, there are only a few, most often simpler approaches for loss estimation in rural areas. The loss assessment can provide an estimate of the financial provisions required for the farmers’ compensation (e.g., in the context of cost- benefit analyses of detention measures). Flood risk is a combination of potential damage and probability of flooding. Losses in agricultural areas exhibit a strong seasonal pattern, and the flooding probability also has a seasonal variation. In the present study, flood risk is assessed for a planned detention area alongside the Elbe River in Germany based on two loss and probability estimation approaches of different time frames, namely a monthly and an annual approach. The results show that the overall potential damage in the proposed detention area amounts to approximately 40000 € a-1, with approximately equal losses for each of the main land uses, agriculture and road infrastructure. A sensitivity analysis showed that the probability of flooding (i.e., the frequency of operation of the detention area) has the largest impact on the overall flood risk.

Published as: Förster S, Kuhlmann B, Lindenschmidt K-E and Bronstert A. 2008. Assessing flood risk for a rural detention area. Nat. Hazards Earth Syst. Sci., 8, 311–322.

74 Assessing flood risk for a rural detention area

1.1 Damage estimation methods This study estimates losses associated with the 1 INTRODUCTION flooding of a detention area in a rural environment. The review of flood loss estimation meth- Flood risk management measures aim to re- ods, therefore, focuses on floodwater damage to duce the negative effects of floods. The designa- croplands and grasslands and road infrastructure, tion of detention areas as one of these measures which are typical land-use types in such flood de- is currently being discussed for the Elbe and tention areas. many other rivers. Several sites along the middle Flood damage estimation methodologies are course of the Elbe River (Germany) have already applied in many countries in Europe (Meyer and been proposed as potential locations for flood Messner, 2005) and worldwide (Dutta et al., detention, and were investigated in terms of 2003). These methods are useful in conducting flood peak reduction potential (IKSE, 2003; cost-benefit analyses of the economic feasibility Helms et al., 2002). However, stakeholders, such of flood control measures. In Germany, respon- as farmers, are reluctant to allow allocation of ag- sibility for flood policy lies with the individual ricultural lands for flood detention, because of federal states and, hence, there are large differ- the negative effects inundated waters have on ag- ences in the character and application of flood ricultural lands (crop losses, excessive sediment estimation methods in these states. The investi- and contaminant deposition, potential degrada- gated site is located in the federal state of Saxony- tion of the soil, etc.). In order to provide decision Anhalt, where damage evaluation is still rarely support for this controversial debate, it is neces- used, according to Meyer and Messner (2005). sary to have an in-depth assessment of the flood However, with the implementation of the new risk of the proposed sites. European Directive on flood risk management The objective of the present study is to inves- (EU, 2007) and the increasing availability of data, tigate the effect of time-varying damage in the it is expected that damage evaluation will gain flood risk assessment of rural flood prone areas. more importance in flood defence planning in The concept is tested at a proposed flood deten- the coming years. tion area at the Elbe River. Section 1 gives a Expected losses in rural areas are typically short overview of flood loss estimation methods much lower than those in urban areas. Hence, with a focus on rural damage. It shows that agri- damage evaluation in rural areas is often ne- cultural losses have a strong seasonal variation, glected or only accounted for by using simple while the flooding probability also varies with approaches and rough estimates. seasons. In order to account for this variability, Pivot et al. (2002) differentiate between losses flood risk that is to be expected for the detention due to damage to crops grown at the time of area is assessed based on two loss and probability flooding, and damage affecting soil characteris- estimation approaches of different time frames, tics. The first is mainly due to the anoxia suffered namely a monthly and an annual approach (sec- by the crop, the water column pressure and lo- tion 2). During the large Elbe flood in August cally the flow of the water. It results in a reduc- 2002, an area of 200 km2 on the right side of the tion in yield and crop quality and may require ad- Elbe River including the proposed detention site ditional expenditures for sowing, tillage, and the was flooded due to several dike failures (BfG, application of fertiliser and crop protective 2002). This flood event enables a validation of agents. The second refers to a potential decrease the damage estimation methods using damages in the quality of soil due to pollutant deposition recorded at the municipal level (section 3). In a and a loss of soil structure due to compaction or sensitivity analysis the relative importance of the erosion. factors crop share, market price and probability Main variables that define the flood damage to of polder operation were investigated (section 4). agricultural lands are the time of year of flood Finally, the two different flood loss estimation occurrence, water depth, duration of flooding, methods, their applicability in other locations, flow velocity, and deposition of pollutants and the potential impact of future developments (DVWK, 1985; LfL, 2005; Citeau, 2003). Many (i.e., land-use changes, frequency of polder opera- authors point out that the time of occurrence of a tion) on the results are discussed. flood with respect to crop growth stages and critical field operations plays a crucial role in the magnitude of damage (Penning-Rowsell et al., Chapter VI 75

2003; Todorovic and Woolhiser, 1972). This dif- ing of winter wheat but necessitates additional fers significantly from damage evaluation in other tillage operations. Generally, loss estimates damage categories, for example damage to build- should be developed for each crop type and pe- ings where loss potential does not vary with the riod of flooding, making allowance for yield seasons. For example, flooding in June/July re- losses due to delayed planting, replanting costs, sults in much higher losses for summer grain savings due to costs not incurred, and costs for crops just prior to harvesting than flooding in clean-up. August just after harvesting. Depending on the Table 1 summarises the agricultural damage time of flooding and the affected crop types, the variables that have been accounted for in selected farmers may decide to undertake measures in or- case studies of flood damage estimation. In most der to alleviate overall loss. USDA (1978) lists case studies, time of occurrence is considered measures to alleviate flood losses depending on whereas the flood variables water depth, inunda- the time of year categorised in half-month peri- tion duration, and flow velocities are only in- ods for pasture and several crop types. For ex- cluded in a few case studies. This is because the ample, it may be possible to replant winter wheat data needed to quantify the impact of these vari- in October with no or low yield reduction, ables on the expected damage are sparse. Citeau whereas it may be too late for replanting in No- (2003) gives a rough estimate of maximum toler- vember, the only option being to plant a substi- able submersion time, inundation depth, and tute spring crop. This saves costs for the harvest- flow velocity for different rural land-use types.

Tab. 1: Comparison of case studies on flood damage estimation including agriculture losses regarding the considered flood variables. Reference Submersion Water depth Submersion Flow velocity (case study site) period duration Hoes and Schuurmans, 2005 no stage-damage no no (The Netherlands) curve Neubert and Thiel, 2004 yes (four periods no no no (Gemany) per year) Dutta et al., 2003 (Japan) yes (monthly) yes (three duration-damage no classes) curve Citeau, 2003 (France) yes (monthly) yes (three yes (four classes) yes (three classes) classes) Consuegra et al., 1995 yes (15-day period) no yes (two classes) no (Switzerland)

The maximum tolerable levels refer to the Another variable causing agricultural losses is conditions that plants are expected to withstand the deposition of waste and mud that might con- without severe damage. The estimates were de- tain pollutants. Such losses often necessitate addi- rived from a survey among farmers in France. tional clean-up costs, and the inundated crops According to Citeau (2003), maximum tolerable and vegetables may not be sold due to contami- inundation duration for cropland varies between nation. three days in spring/summer to one month in au- Other agricultural goods that may be suscepti- tumn/winter. Maximum tolerable depth of sub- ble to flood damages are farm buildings, machin- mersion strongly depends on the type of land use ery, and infrastructure (e.g. roads). In contrast to and vegetation height. Examples provided in crop and grassland losses, damage in these other Citeau (2003) are 1 m for orchards and 0.5 m for categories is independent of the season. Usually vineyards. Maximum flow velocities vary between stage-damage functions are applied which relate 0.25 m s-1 for field vegetables and 0.5 m s-1 for the water level to the relative expected damage. orchards. No velocity values are provided for In order to obtain an estimate of the total ex- cropland. High flow velocities can cause direct pected loss, the relative damage is related to the damage to the plants and to soil degradation maximum damage per area and land-use type from erosion (LfL, 2005). (Merz et al., 2004). Indirect losses due to traffic 76 Assessing flood risk for a rural detention area and business interruptions are usually estimated terval. The storage capacity is 40 million m³. The as a proportion of direct costs (YRFCMP, 2003). detention area consists of agricultural land with very fertile soils and high agricultural productiv- ity. More than 90% of the land is currently under 1.2 Study site intensive agricultural use. The remaining 10% of The present-day embankments confining most the area consists of watercourses and forest. of the German reaches of the Elbe River date There are no settlements within the proposed de- back to the 2nd half of the 19th century, al- tention area. It is expected that the area will retain though dike construction along the Elbe began as its present function as agricultural land even after early as the 12th century. The embankments have it has been designated as a detention area. led to a reduction of the retention area in Ger- In order to estimate the expected flood losses many from 6172 km2 to 838 km2 (13.6% of on agricultural lands in the detention area, infor- original). The reduction of retention areas and mation is needed on the type and mixture of the straightening of the main river channel have crops typically grown on those lands. The agricul- resulted in an acceleration of flow velocity and an tural land-use types for the years 2002 to 2007 increase of the flood water levels (BfG, 2002). were collected by interviewing the local farmers. Today the construction of detention sites in The farmers’ decision about which crops to grow the former inundation area along the Elbe is be- depends on the profit margins for different crops ing discussed. Such sites would enable controlled and the farmers’ goal, which is assumed to be diversion and storage of excess water during large profit maximization. Grain crops are grown on flood events in order to reduce flood risk adja- 51% of the agricultural area due to the good soil cent to and downstream from the detention ar- quality in the former inundation area (Figure 2). eas. Main grain crops grown in the study area are wheat and barley. Corn (9%) is used for energy production and silage fodder. The share of grassland is comparatively low (7%). Grass is usually cut three times per year and is mainly used for fodder production.

Fig. 1: Detention area with agricultural fields and road sys- Fig. 2: Land use of the study site (% of agricultural land) tem. It is confined by the Elbe main dike to the river and polder dikes to the hinterland. 2 METHODOLOGY In the present study, one large controlled detention area was investigated that is already in the Risk is defined as the probability of the ad- early planning stages (Figure 1). It is situated verse effects of a natural process, such as a flood, alongside the right bank of the middle course of exceeding a certain magnitude (intensity) from the Elbe River between the Torgau and Witten- which certain damages and losses occur (vulner- berg gauges and is designed for reducing flood ability) (Merz et al., 2007). For the detention area, peaks having a 100-years or more recurrence in- Chapter VI 77 the probability of flooding corresponds to the subdivided politically into 38 administrative re- probability of opening the inlet gate for flood wa- gions, each of which has different MV values for ter diversion, which would be the case for large each crop. The MV values for each region were floods with return periods exceeding 100 years. derived from the standard gross margins pro- The costs are associated with the flood losses on vided by the Curatorship for Technology and agricultural land and the road system within the Construction Engineering in Agriculture (KTBL, detention area. Since loss on agricultural land has 2007). The MV values for the administrative re- a seasonal variation, the flood frequency analysis gion of Dessau/Saxony-Anhalt, in which the provides monthly weights on the flooding prob- study site lies, are given in Table 2 for selected ability. The annual monetary flood loss in € per crops. hectare per year (€ ha-1 a-1) on agricultural fields The damage impact factor DI depends on the is calculated by weighting the loss from a single type of crop, the month of the flooding occur- flood event occurring in each month with the rence, and the inundation duration. Table 3 gives probability of flooding in that month (Hess and an example of damage impact percentages for Morris, 1987). wheat and grass for each month. The information is based on empirical data from surveys in France and Germany as referenced in LfUG 2.1 Damage estimation (2005) and expert knowledge. The damage im- Several approaches of varying complexity are pact factors can reach values of up to 100% indi- available to calculate losses in agricultural pro- cating a total loss. The impact is particularly de- duction due to flooding. This study applies two pendent on the growth stage of each crop. Root approaches, using a monthly and an annual time crops and grain crops are harvested once per year frame for loss estimation. and their impact factors have patterns similar to A damage estimation model based on a the ones shown for wheat. Their impact factors monthly disaggregation of damages to crops and are differentiated into four classes of inundation grasslands was developed within the framework duration. Grass is an exception to the other crops of the project “Methods for the evaluation of di- because it can usually be harvested three times rect and indirect flood losses” (MEDIS, 2007). per year (May, July, and August). The total annual The expected damage for each crop is calculated yield of grass is distributed throughout the year in by: three harvests in May, July, and August with an 12 annual average of 50%, 20% and 30%, respec-  m  DIPMMVED m (1) tively. Hence, the impact factors are lower since m1 only a fraction of the total yearly harvest is dam- where ED = expected damages (monetary aged by a flood. The impact factors are also in- losses in € ha-1 a-1), MV = market value (that can dependent of inundation duration because sedi- be obtained by the harvested crop without flood- ment deposition on grasslands occurs after every ing in € ha-1), PM = probability of polder flood- flood, regardless how short the inundation period ing every 100 years for a certain month m (a-1) is, making the grass unusable for high value fod- and DI = damage impact on crops for month m der. For inundation times longer than about 10 (%). The market value MV is calculated by the days, additional costs may be incurred due to total yield of a crop harvested multiplied by its structural damage to the grass roots requiring a selling price. MV differs from region to region repair seeding of the grasslands. The costs for the since the crop yield is dependent on the climatic repair seeding of grasslands, which includes and soil conditions and the type of agricultural seeds, labour and machinery are approximately management practices used. Germany can be 45 € ha-1 (KTBL, 2006).

Tab. 2: Market value of selected crops for the administrative region of Dessau/Saxony-Anhalt averaged over the years 2000 to 2005. crop wheat rye barley corn canola potatoes sugar beets grassland market value (€/ha) 704 459 605 883 632 2339 2103 266

78 Assessing flood risk for a rural detention area

Tab. 3: Damage impact factors for wheat and grass for different months of the year grouped by different durations of flooding. Values have been extended from LfUG (2005). Wheat Grassland inundation 1-3 days (%) 4-7 days (%) 8-11 days (%) > 11 days (%) 1-11 days (%) duration January 5 10 20 80 5 February51020805 March 5 10 20 80 10 April 10 25 40 80 20 May 20 40 70 100 50 June 50 50 80 100 15 July 100 100 100 100 20 August 100 100 100 100 30 September 0 0 0 0 10 October 5 10 20 80 10 November 5 10 20 80 10 December 5 10 20 80 10

Figure 3 shows the expected damage for each flooding occurs was not considered. Damages crop differentiated into classes of inundation du- with the annual approach are calculated by: ration. The maximum damage is expected to vary between 10 and 16 € ha-1 a-1 for grain crops and    PARDMVED (2) between 32 and 36 € ha-1 a-1 for root crops based on an inundation duration of more than 11 days. where ED = expected damages (monetary Damages for grass are the lowest at approxi- losses in € ha-1 a-1), market value (that can be ob- mately 1 € ha-1 a-1. tained from the agricultural land without flooding In addition to the monthly damage estimation in € ha-1), RD = relative damage costs (%) and model, an annual approach was applied in which PA = probability of polder flooding every 100 only two land-use classes were distinguished and years (i.e. 0.01 a-1). the time within the growing season when the

Fig. 3: Expected damages to grain crops (wheat, rye, barley, corn), oilseed plants (canola), root crops (potatoes and sugar beets) and grass based on flooding occurrence categorised on a monthly basis. Data are derived from LfUG (2005), KTBL (2006) and KTBL (2007). It is assumed that the inundation duration of > 11 days classification corresponds to the degree of damage expected to occur within the polders. Chapter VI 79

The agricultural land was differentiated in ar- The expected damage to the road system was able land and grassland with market values of then calculated by: 4000 and 2000 € ha-1, respectively. These figures are based on damage claims from past extreme    PARDRCED (3) flood events in the state of Saxony in Germany (LfUG, 2005).The relative damages to both, re- where ED = expected damages (monetary gardless of flood depth, inundation duration or losses in € ha-1 a-1), RC = replacement costs the time within the growing season, were set to (€ ha-1), RD = relative damage (%) and be 50% and 10%, respectively. According to PA = probability of polder flooding every 100 LfUG (2005), these values were found to fit re- years (i.e. 0.01 a-1). Based on damages recorded corded damages in flat inundation areas best. In during past flood events, damage to the traffic comparison, relative damages in mountainous ar- system is given as 200 € m-2, whereas a relative eas with discharges of 1 m2 s-1 increase to 75% damage impact factor of 10% is provided for wa- and 25%, respectively. The damage to be ex- ter depths larger than 1 m and flow velocities be- pected when operating the polders to cap floods low 1 m s-1 (LfUG, 2005). This corresponds that exceed discharges with return periods of closely to repair costs of 25 € m-2 for asphalt more than 100 years are 20 € ha-1 a-1 for arable roads that were found on the basis of bid prices farmland and 2 € ha-1 a-1 for grassland. These after the deliberate flooding of a polder system damage values are of the same order of magni- further downstream along the Elbe River (Ell- tude as the damages calculated on a monthly ba- mann and Schulze, 2004). For a probability of sis (compare Figure 3). flooding of 1%, which corresponds to the opera- In order to provide a representative picture of tion of the polders every 100 years, the expected the current land-use situation, including crop ro- damages would amount to 2000 € ha-1 a-1. Al- tation schemes, the percentage shares of crop though this value is high compared to the ex- types and grassland were averaged over the last 5 pected damages for arable land and grassland ob- years (2003-2007). For the annual approach, this tained with the annual approach, the total road information was aggregated into two classes of surface area is substantially less than that taken arable land and grassland. up by agricultural fields. During the large flood in August 2002 having In the present study loss estimation is re- a return period of approximately 180 years near stricted to direct tangible damage. Indirect dam- the study site the region where the proposed de- age such as traffic interruption is considered to tention basin would be located was flooded as a be relatively small in the rural study area. Since result of dike failures. Afterwards losses were re- the detention area is not inhabited and the people corded by the authorities for compensation pur- will be warned prior to the polder operation, no pose. Economic loss information for agricultural victims or loss of livestock is expected. Intangible land on the municipal level was made available damage is mainly expected in the form of adverse for the present study. In order to assess the qual- impacts on flora, fauna, and the terrestrial and ity of the results, recorded loss data for one mu- aquatic ecosystem in the affected area. In particu- nicipality were compared with estimated losses lar, the water quality degradation from flooding for the same municipality. The municipality was can have negative effects on the fish fauna as re- chosen because it has a share of the proposed de- ported in studies on storage basins and flood- tention area and was almost completely inun- plains (Knösche, 2003; Howitt et al., 2007). This dated in 2002, as indicated by satellite imagery. aspect will be part of future work on the same Settlements were less affected because they are detention area. built on slightly higher elevated ground. Analo- gous to the detention area, data on the percent- 2.2 Flood frequency analysis age of crop types and grassland in the selected municipality were collected. As the costs associated with flooding of agri- Besides losses in the agricultural sector, infra- cultural land are differentiated on a monthly ba- structure damage in the form of damage to the sis, the expected percentage distribution of dam- road system is considered to be the other major aging floods was also analysed monthly using the damage component in the study area. Informa- discharge recorded from the gauge at Torgau for tion on length and width of the roads was col- the time period 1936 – 2004. The Torgau gauge lected from aerial photographs and field surveys. is located approximately 30 km upstream of the 80 Assessing flood risk for a rural detention area proposed detention area. There are no relevant cally larger. This indicates that there is a relation tributaries on the river stretch between the gauge between seasonality of floods and their magni- and the detention site. Figure 4 shows the tude, which should be accounted for when de- monthly distribution of all flood peaks in the an- termining the monthly percentage distribution of nual maximum series (AMS) and of the largest damaging floods. The differentiation into months 10% of the AMS flood events. 73% of the flood having different flooding pattern is motivated by events in the AMS occur during the hydrological the strong dependence of losses on the month of winter season from November to April (with flood occurrence. In Figure 4, the damage impact more than 30% of the events occurring in factors for wheat are included to illustrate this March). July and August events constitute 12% aspect. Extreme flood events with peak dis- of the AMS events, however account for 27% of charges relevant for polder operation have a high the largest 10% of the AMS events. Apparently, probability of occurrence during the summer there are many AMS events with comparatively months shortly before harvest, when grain crops small peak discharge values in spring, whereas in are most vulnerable to inundation. summer AMS events are less frequent, but typi-

Fig. 4: Seasonality of annual peak flows for all and the 10% largest events based on the discharge AMS for 1936-2004 at the Torgau gauge and seasonality of the damage impact factor for wheat for inundation durations of 8 to 11 days.

The seasonality of flood magnitudes is a result (2005) propose a method to isolate the contribu- of different flood generating mechanisms that are tions of individual months or seasons to the an- often dominant during different seasons (Lecce, nual flood frequency curve to account for the in- 2000). If this is the case it is advisable to separate tra-annual variability in flood processes. the flood series into seasons of similar generation Since the dikes are designed to retain floods mechanisms. Petrow et al. (2007) investigated the with return periods of up to 100 years and hence relation between dominate European atmos- the detention area is operated only during very pheric circulation patterns and annual maximum large flood events, it is necessary to determine the flood events for a sub-catchment of the Elbe ba- probability that this discharge will be exceeded. sin. They found that westerly and north-westerly From the Torgau gauge discharge record for the cyclones are responsible for most winter floods, years 1936 – 2004, the largest flood in the entire but only play an important role for return periods year and in each of the 12 months is picked to up to 10 years. Larger floods with return periods construct annual and monthly flood frequency larger than 50 years are exclusively generated by a curves (Figure 5). A composite of the GEV Vb-weather regime, which is characterised by a (Generalised Extreme Value) (Kotz and Nadara- cyclone system travelling northeastward from the jah, 2000) and GL (Generalised Logistics) (John- Mediterranean to Central Europe. Sivapalan et al. son et al., 1994) distributions using L-moments Chapter VI 81 gave the best fit to the data. Both distribution corresponding to the discharge of 4000 m3 s-1 are functions are widely used in flood frequency larger (for example about 150 years for March), analysis. The composite distribution function is a i.e. the occurrence probabilities smaller. This combination of the two functions, which were means that the probability of a flood peak of a given equal weights (Merz and Thieken, 2005). certain discharge (for example 4000 m3 s-1) oc- Figure 5 shows that the discharge associated with curring in a particular month is smaller than its the annual return period of 100 years is probability of occurrence at any time of the year. 4000 m3 s-1, while the monthly return periods

Fig. 5: Flood frequency analyses based on the annual and monthly maximum discharges of the years 1936 – 2004 at the gauge at Torgau. A composite of the GEV and GL distributions is used.

undated for more than one week. Maximum flow velocities of 1.4 m s-1 are simulated behind 3 RESULTS the inlet gate. Areas with maximum flow velocities of more than 1 m s-1 are restricted to the The temporal and spatial distribution of stilling basin behind the inlet gate and along an flooding variables, such as inundation duration, already existing stream through the detention water depth, and flow velocity were obtained in area. Figures 6 and 7 show the spatial distribu- previous 2D-hydrodynamic simulations of the tion of the inundation duration and the maxi- same detention site based on the large flood mum flow velocity in the detention area, re- event of August 2002 (Förster et al., 2008; spectively. The results are based on the 2002 Chatterjee et al., 2008; Huang et al., 2007). flood event, which was characterised by a rather Simulated water depths in the detention area steep flood hydrograph. Inundation duration is range from 0.5 m in the higher elevated south- expected to increase for flood events having ern part to 5.7 m in the central part, with a wider flood hydrographs than the 2002 event, mean water depth of 2.5 m. The entire deten- because the emptying process will start not ear- tion area remains inundated for three days until lier than the Elbe water level falls below a level the start of the emptying process. After day that allows for safe discharge at the down- four, the surface water retreats from only 5% of stream river reaches the area, whereas 75% of the area remains in- 82 Assessing flood risk for a rural detention area

Fig. 6: Simulation results for the flood event of August Fig. 7: Simulation results for the flood event of August 2002 (inundation duration in days). 2002 (maximum flow velocity in m s-1).

The farmers interviewed stated that most mated loss to the road infrastructure, overall an- fields were not accessible for machines for sev- nual damage obtained with the monthly and an- eral weeks or even months after the August 2002 nual approaches ranges between 46000 € a-1 and flood due to high soil moisture and sludge depo- 37000 € a-1, respectively. The negative effects on sition, although the surface flood water had long the total production process of the farming op- retreated. Hence, the case of more than 11 days eration (e.g., reduction in animal production from may realistically represent the agricultural damage diminished fodder quality, changes in crop rota- and was applied in the estimation of the annual tion, non-fulfillment of delivery contracts) were damage using the monthly approach. not considered due to the difficulty in quantifying The estimated annual damage in agricultural these effects on a regional scale. fields for the monthly and annual approach In order to assess the quality of the damage amounts to 21400 € a-1 and 14600 € a-1, respec- estimation methods, recorded and estimated tively. The annual damage to the road system was losses for one representative municipality were estimated to be 15800 € a-1. Apart from damages compared. The agricultural losses recorded for to field crops, additional losses to agricultural this municipality for the specific flood event of production due to damages to buildings, machin- August 2002 amounted to 644000 €. Losses in € ery, inventory, and clean-up measures occur. were estimated with the monthly and annual ap- They are very site-specific and not easy to esti- proaches for the same flood event. As these are mate. The loss information collected by the au- event values, annual or monthly flooding prob- thorities for the affected municipalities during the abilities were not considered. Estimated losses flood in 2002 gives an indication of the magni- using the annual approach are much higher tude of these additional losses. An average of (3569000 €) than those using the monthly ap- 11% for building damages, 3% for machinery proach (546000 €). This is because in the annual losses, 7% for inventory losses, and 12% for approach, the damage values are independent of clean-up costs out of the overall agricultural when the flood occurs during the growing season losses in the flood affected area was recorded. (April-October) and therefore constitute an aver- Together, they make up approximately 30% of age of the expected losses. In the monthly ap- the overall agricultural losses. Adding these addi- proach damage impact factors were applied ac- tional costs to the loss estimates obtained with cording to the specific month in which the flood the monthly and annual approaches results in occurred. At the time of flooding at the end of overall losses of approximately 30500 € a-1 and August, most of the cereal fields had already been 21000 € a-1, respectively. Together with the esti- harvested and, hence, estimated losses were com- Chapter VI 83 paratively low. It illustrates the impact that the - environment – all of the land is converted to time of flood occurrence has on the overall loss. grasslands (grass has a lower oxygen demand Depending on the time of occurrence, the ex- on overlying flood waters than do tilled fields pected agricultural losses associated with a spe- and, hence, adverse ecological effects, such as cific flood event in the detention area vary be- stress on fish populations due to oxygen defi- tween 287000 € in January and 994000 € in July. ciency, will be reduced).

The actual crop production will be a mixture 4 SENSITIVITY ANALYSIS of the scenarios. Figure 8 shows the results of the four scenarios compared to the current land use A sensitivity analysis was performed to deter- in the detention area derived using the monthly mine the relative importance of different factors damage estimation approach. It is evident that that are directly influenced by humans. The fac- grains, canola and corn (grain crops and energy sce- tors included: narios) do not change the expected damages significantly from the current situation because a 1. crop share of agricultural land use majority of the land coverage is currently a mix 2. market price for crop types (± 20%) of these crops. However, focusing on the pro- 3. probability of polder operation (HQ80) duction of root plants (root crops scenario) would increase damages by 2½ times. In comparison, To account for the sensitivity of the results to expected damages to grasslands (environment sce- different crop shares, four land-use scenarios nario) are minute. were considered, which involved allocating the Changing the market price of the crops in the entire land coverage of the polder area to either current situation by ±20% would vary the ex- grain crops, root crops, energy plants, or grass- pected damages by the polder operation by ap- lands: proximately the same degree (17% increase and 22% decrease in expected damages if the crop - grain crops – 100% grain crops (wheat, rye, price is increased or decreased by 20%, respec- barley) tively). The probability proved to be a sensitive - root crops – 100% root crops (potatoes, sugar factor with expected damages doubling if the beets) polders were to cap discharge peaks of flood - energy – 100% of crops used for biomass en- events having a return period of 80 years (i.e., 3 -1 ergy production or as biofuels (corn, canola) Qpeak = 3300 m s as opposed to 100 years with 3 -1 Qpeak = 4000 m s ).

Fig. 8: Sensitivity of land use, market price and probability of polder operation on the expected damage. 84 Assessing flood risk for a rural detention area

proach, damage impact factors often reach 100% for arable land because of the long inundation 5 DISCUSSION AND CONCLUSIONS times that are characteristic for detention areas, whereas in the annual approach a uniform dam- Although agricultural damage is often low age impact factor of 50% is assumed for arable compared with urban or infrastructure damage, it land. Depending on the specific characteristics of should be accounted for in areas where agricul- flood prone area with respect to the shares of tural production is a predominant activity (Mess- land-use types and the pattern of flooding prob- ner et al., 2007). The proposed monthly damage ability both approaches may result in different assessment procedure is applicable to a wide va- risk assessments. The monthly approach is more riety of agricultural schemes that are character- desirable as it is likely to provide more accurate ised by seasonal variation in plant growth and estimates. Advantages of the annual approach hence expected losses due to flooding. are, however, the low data requirements and a The damage to agricultural production that re- less time-consuming estimation procedure. sults from flooding during a specific flood event If losses for certain flood events in € instead mainly depends on the time of occurrence rela- of annual damages in € per year are to be esti- tive to the growth stages and the share of crop mated, the monthly approach seems even more types and grassland in the area flooded. Unfortu- adequate, since the loss estimates strongly depend nately, bibliographic sources only provide little on the flood occurrence time. The example of information on the resistance of crops to floods the municipality that was flooded in August 2002 (Citeau, 2003). The market value as a product of demonstrated the large differences in estimated total yield and selling price varies greatly between losses between both approaches. the different agricultural land-use types, while Estimated losses to the road system also con- each type exhibits a different seasonal pattern of stitute a large proportion of the overall expected expected losses. losses. Damage potential to the road system in Other damage variables, such as water depth, the study area has even increased over the past inundation duration, and flow velocity, are less years. This is because after the extensive inunda- relevant in case of flooding of an agriculturally- tion of the area during the flood event in summer used detention area. This is due to the fact that 2002, several formerly unpaved field lanes were the water depths are comparatively high in order reconstructed with an asphalt cover that bears to provide a large storage volume compared to larger reconstruction costs in case of future in- the ground surface area. In most cases, a total undations. yield loss has to be assumed because of the com- The sensitivity analysis showed that in flood bined adverse effect of damages and the re- risk assessments of rural areas with low intensive stricted accessibility after the flooding due to land use it is more important to evaluate the high soil wetness. The operation of detention ar- variation in flooding probability than the varia- eas is a special case of inundation in the sense tion in land use. It is particularly of importance as that the flooding occurs deliberately with warning large summer floods are becoming more likely to times long enough to undertake measures that al- occur. According to Kundzewicz et al. (2005), leviate the losses, such as bringing in the harvest, projected increases in temperature and associated evacuating livestock, or removing machinery increases in potential water content and intense from the flood prone area. precipitation are expected to increase summer The applied monthly and annual approaches flooding in most of Central Europe. Not only the are based on market values of the grown crops in flood magnitude, but also the seasonal distribu- order to estimate agricultural production losses, tion of flood occurrence is likely to be affected whereas damages to farm buildings, machinery by climate change (Sivapalan et al., 2005). Deten- and inventory as well as clean-up costs were not tion basins and other flood management meas- considered. Hence, both damage results are com- ures are one option to cope with future changes parable. The comparatively lower estimated an- in flooding probability. nual damages obtained with the annual approach can be explained by the specific conditions in the study area. The fertile soils allow high yields from the intensive production of crops with high market prices. Furthermore, in the monthly ap- Chapter VI 85

ACKNOWLEDGEMENTS paper reflects the authors’ views and not those of the European Community. Neither the European The research was jointly funded by the Ger- Community nor any member of the FLOODsite man Ministry of Education and Research Consortium is liable for any use of the informa- (BMBF) within the framework of the project tion in this paper. MEDIS - Methods for the Evaluation of Direct The authors are grateful to the Landesbetrieb and Indirect Flood Losses (No. 0330688), the für Hochwasserschutz und Wasserwirtschaft Germany Research Foundation (Helmholtz Sachsen-Anhalt and the Wasser- und Schiff- Young Scientists Group “Information and mod- fahrtsamt Dresden for data provision. Dr. Sibylle eling systems for large scale flood situations”) Itzerott is thanked for her company und support and the Sixth Framework Program of the Euro- on the field trip. pean Commission (FLOODsite project, EC Con- tract number: GOCE-CT-2004-505420). This

Chapter VII

Conclusions and recommendations

In this thesis, hydraulic, environmental and Table 1 compares the ratio of storage capacity economic impacts of flood polder management to basin size and storage capacity to mean an- were investigated using model approaches of nual flood discharge for several flood polders in different complexities. The research focused on Germany including the two flood polder sys- two flood polder systems built for the mitiga- tems that were investigated in this thesis. The tion of the flood hazard on the Middle Elbe higher the two ratios, the larger is the general River in Germany. flood peak reduction potential. However, spe- Based on the research questions posed at the cific conditions such as flood predictability and beginning of the thesis, this chapter summarises polder topography have to be considered when and discusses the research results and gives rec- evaluating the effectiveness of each single site. ommendations for flood polder management and future research. Tab. 1: Comparison of storage indicators of several flood polders in Germany Flood polder Storage Storage Storage 1 CONCLUSIONS capacity capacity to capacity to basin size MHQ* (million m³) (m³/km²) (m³/(m³/s)) 1 18.0 358 5788 1.1 Hydrodynamic modelling Altenheim / River Rhine

Research Question: How effective are the Riedensheim2/ 8.3 415 7155 flood polders in terms of peak reduction for Danube River varying flood scenarios and operational (planned) schemes? Axien3/ Elbe 40.0 724 28369 Hydrodynamic models were applied in both River (planned) study areas to investigate the peak reduction under different flood scenarios and operational Havelpolder4/ 110.0** 1125 62147 schemes. The study demonstrated that flood Elbe River peak levels can effectively be capped by tempo- Rösa5/ Mulde 21.5 3484 44792 rarily storing excess floodwater in flood pol- River (under ders. However, it also showed that the obtain- construction) able flood peak reduction strongly depends on * MHQ = mean annual flood discharge the shape of the flood hydrograph. The event- ** only Havel flood polders without Havel floodplain specific shape is related to catchment characteristics such as basin size, soil permeability or Gauge and discharge time series used: 1 gauge Maxau tributary pattern and flood event characteristics 1931-2003 (BfG, 2003), 2 gauge Ingolstadt 1975-2001 such as rainfall pattern and initial soil moisture. (BfG, 2001), 3 gauge Torgau 1935-2004 (BfG, 2004), 4 gauge Tangermünde 1961-2002 (BfG, 2002a), 5 gauge In order to assess the potential effectiveness Bad Düben 1961-2006 (LfUG, 2006) of flood polders in terms of peak reduction, 88 Conclusions and recommendations

The timing of opening a flood polder is cru- which is home to several protected species and cial for a successful operation as pointed out by put under protection according to national and several authors (Galbáts, 2006; Rátky and European nature conservation legislation. Szlávik, 2001). If the storage is utilised too Numerical water quality simulations in the early, most of the detained volume is taken flood polder Axien showed that under the con- from the rising limb. If the inlet gate is opened ditions used in this study oxygen concentration too late, the volume is merely taken from the in the flood polder falls below 3 mg l-1. This le- falling limb. In both cases the peak lowering vel is considered critical for fish (Böhme et al., obtained will be less than potentially possible 2005), while Wolter et al. (2003) point out that (Silva et al., 2004). In this study the opening there are large variations among the species re- time of the inlet gate was pre-determined for garding tolerance levels and lethal oxygen con- each investigated flood scenario so as to maxi- centrations. The low simulated oxygen levels mise flood peak reduction. can mainly be attributed to the low flow veloci- Dividing the inlet gate into separately oper- ties and large amount of degradable organic able parts has been proved to be a promising matter in the flood polders compared to the strategy for an adjusted gate control. It also al- river, leading to a strong de-oxygenation of the lows for the utilisation of the flood polder in flood water. Warm and calm weather condi- case of a gate malfunction. tions, as observed shortly after the August 2002 Flooding characteristics as obtained from flood event, will even intensify the depletion of the one-/two-dimensional hydrodynamic dissolved oxygen. Adverse impacts on the water model showed a large spatial and temporal quality of the Elbe River itself when releasing variation in flow velocity and water depth the oxygen-poor stored water are negligible due within the investigated flood polder Axien. to the high Elbe discharge compared to the These variables are relevant in the subsequent much smaller outflow from flood polder Axien water quality model, which was coupled to the and Lower Havel River, respectively. hydrodynamic model. 1.3 Vulnerability assessment

1.2 Water quality modelling Research question: How can economic vul- Research question: How does the long water nerability be assessed when considering time- storage affect dissolved oxygen levels in the varying damage in agricultural areas? flood polders? Both study areas are characterised by an in- During the operation of the Havelpolder tensive agricultural land use with grain crops, system in August 2002, a large number of fish corn, canola and grassland being the main crop died due to a considerable de-oxygenation of types. Agricultural fields show a typical growth the water in the flood polder reservoirs and the pattern throughout the year. Losses in case of a subsequent release of stored oxygen-poor water flooding therefore strongly depend on the time into the Lower Havel River which is known for of flood occurrence relative to growth stages its wealth of fish (Knösche, 2003). An envi- and agricultural field operations. Furthermore, ronmental risk for fish and other aquatic ani- losses vary greatly among the crop types. mals may occur if oxygen-poor flood water is Not only the vulnerability of agricultural released to a river reach with a small ratio of land, but also the flooding probability varies storage capacity to mean annual flood discharge with seasons. Often the seasonal flooding prob- or if the flood polder itself contains aquatic ability is different for smaller floods than for habitats. The first applies to the Havelpolder very large floods which are relevant for flood system, where water is released to the Lower polder operation. Extremely large floods of the Havel River, which is a comparatively small Elbe River have a higher occurrence probability tributary of the Elbe River with a mean annual during summer when agricultural vulnerability flood discharge of 212 m³/s (based on the dis- is highest, compared to lower floods which are charge series 1981-2002 at the gauge Havelberg most frequent in spring. To account for the (BfG, 2002b)). The second applies to the seasonal variability in both expected agricultural planned flood polder Axien, where a small losses and flooding probability, a monthly dif- stream runs through the flood polder area, ferentiation was chosen in this study. Chapter VII 89

Different from the time of occurrence, agri- ence in relative losses is mainly attributed to cultural losses in the flood polders do not vary there being less vulnerable land use in the much between floods of different magnitude Havelpolder system, in particular the propor- and duration. However, the potential loss in the tion of grassland. While grassland coverage is benefiting areas, i.e. the areas in which the nearly 70 % in the Havelpolder system, it is flood hazard is reduced due to flood polder only 7 % of the agricultural land in the planned utilisation, strongly depends on these flood flood polder Axien. Relative losses also vary characteristics. This is due to the fact that for among the single flood polder reservoirs in the flood polder, the affected land is restricted both study areas as result of prevailing land use to the area within the polder dikes, and damage and storage capacity. to crops is relatively independent of water Apart from losses that occur in the flood depth. In the benefiting area, however, a flood polders during their utilisation, further cost fac- of larger magnitude and duration may result in tors have to be considered when assessing cost- a larger flood-affected area and higher inunda- efficiency of flood polders, however, they were tion depths in areas of various land use types. mostly beyond the scope of this study. These Particularly in urban areas, loss strongly corre- factors include the damage mitigation or re- lates with inundation depth. duced expenses for flood management in the Under the current land use, the expected benefiting areas, possible land acquisition pay- loss per m³ of retained water is estimated for ments or costs for land use conversion and the both study areas (Table 2). Mean expected expenditures for construction and maintenance losses on agricultural lands, including crop of engineering structures such as control gates losses, additional losses due to building and and dikes. Also the flood polder’s location rela- machinery damages and mean losses in the road tive to benefiting areas and their vulnerability infrastructure, are considered. Relative losses must be taken into consideration. Flood polders are significantly higher in the planned flood should preferably be located adjacent to or up- polder Axien than in the Havelpolder system, stream of highly vulnerable areas, but down- although the mean water depth in the Axien stream of flood-relevant inflows into the river. flood polder is considerably higher. The differ-

Tab. 2: Comparison of cost-efficiency indicators for both study areas Case study Losses Storage Losses to storage Area Losses to area Mean water area capacity capacity depth (million €) (million m³) (€/1000 m³) (km²) (million €/km²) (m) Havelpolder 6.7 110 61 100 0.07 1.1 Axien 4.6 40 115 17 0.27 2.4

priate spatial discretisation should be chosen according to the system’s complexity and the 1.4 Model complexity study objective. In this study, hydraulic and wa- Research question: Which is the appropriate ter quality models of different spatial discretisa- model complexity to simulate hydraulic and wa- tion were applied for the simulation of flood ter quality processes in flood polders? polder processes in order to assess the models’ suitability in view of performance and model- Hydraulic processes in flood plains are often ling effort. very complex including three-dimensional flow A one- (1D) and a coupled one-/two- structures (Knight and Shiono, 1996). How- dimensional (1D-2D) model approach were ap- ever, for practical purposes there is a need for plied to simulate the flooding and emptying simplified spatial representations of the model process in the planned flood polder Axien and domain for reasons of limited resources and the flow in the adjacent Elbe River reach. In data availability. All models incorporate certain both model approaches the river and floodplain simplifying assumptions and approximations, geometry was described by a series of cross sec- which pose specific limitations for certain ap- tions. In the coupled 1D-2D approach the plications (Shanahan et al., 1998). The appro- flood polder was represented in the form of a 90 Conclusions and recommendations

Digital Elevation Model (DEM), whereas in the the different approach of determining flow ve- 1D approach the flood polder geometry was locities and water depths in the two models. given solely as a storage function, assuming a The 2D model approach required a signifi- horizontal water surface. cantly higher pre- and post-processing effort Both the 1D and coupled 1D-2D model and longer computation times per simulation simulations yield the same flood wave reduction run. It is therefore not suitable for the investi- in the Elbe River. However, the 1D-2D model gation of various different flood scenarios and provides additional information such as the for testing the model reliability with an exten- spatial variation in flow velocities and water sive sensitivity analysis. However, investigating depths. A two-dimensional simulation of the the impact of the spatial variability within the water flow process in the flood polder may be model domain on the state variables requires a inevitable for certain applications. They include spatially distributed model such as the 2D simu- studying the effects of different land use types lation approach used in this study. on water propagation, identifying flow obstacles Therefore, for practical application it is rec- and preferred flow paths, optimizing the loca- ommended to firstly set up a fast running tion of control structures, gaining information model of lower spatial discretisation like the 0D on the spatial and temporal development of stirred tank model. Secondly, the parameter flow velocities or identifying remaining water sensitivity on the water quality results should be patches during the emptying process of the checked with a particular focus on those pa- flood polder. rameters that are spatially variable and therefore The computational time as well as the stor- assumed to be better presented in a 2D model. age requirements for the 1D model were con- Thirdly, a 2D model should be run if a high siderably lower than for the 1D-2D model and spatial variation in water quality results is as- even more so when finer resolution DEMs are sumed from the previous runs. Generally, 2D used. Further, unlike the 1D model, consider- model approaches may be preferred in flood- able effort was required in setting up and simu- plain areas and flood polders of complex to- lating the 1D-2D model. In view of all these pography and distinct variations in hydraulic factors, it is recommended to use a 1D model variables such as water depth and flow veloci- for studying the flooding processes of polders, ties. particularly the peak reductions in the main river. However, a 1D-2D approach may be used when the study of flow dynamics in the 2 RECOMMENDATIONS polder is of particular interest, if a spatial distribution of the flooding parameters such as inundation duration is required for a subsequent 2.1 Flood polder management damage assessment or if a 2D water quality model is coupled to the 2D hydrodynamic In this section general recommendations and model, as was done in this study. suggestions for flood polder management are A 2D vertically averaged model and a zero- drawn from the investigations. dimensional (0D) stirred tank model were ap- When choosing potential locations for flood plied to simulate the dissolved oxygen dynamics polders the most important criteria are (a) a in the flood polder Axien. The same gate inflow large storage volume compared to the discharge and outflow boundary conditions were used in volume of the river flood wave, (b) suitable to- both models as obtained from the previous 1D- pographic conditions to allow for an efficient 2D hydrodynamic simulation. Water quality filling and emptying, (c) a suitable location, boundary parameters and water quality proc- preferably just upstream or adjacent to the river esses were chosen to be identical in both mod- reaches to be protected, (d) a low vulnerable els, the spatial discretisation being the only dif- land use within the flood polder area, and (e) ference between the two approaches. The the availability of an accurate and timely flood dissolved oxygen dynamics as simulated with forecast to adjust the gate control to the pre- both models were similar in the initial flooding dicted flood wave. period, but deviated with the start of the empty- The utilisation of flood polders is limited to ing process. The deviation can be attributed to the flow time of the flood wave in the framework of the available forecast (LAWA, 1995). Chapter VII 91

The lead time should be long enough to cover ity, the investigations in study area 1 (Havelpol- the whole flood peak to be capped so as to de- der system) demonstrated that it may be advis- termine the optimal gate control strategy with able to restrict release of oxygen-poor water reference to the available storage capacity and into the Lower Havel River with its high fish maximum flow rates through the gates. For op- abundance in order to maintain a sufficient erational reliability, it is necessary to divide the oxygen level in the River. inlet into at least two separately operable parts A controlled emptying process implies the as it allows for the utilisation of the storage area existence of adjustable control structures rather in case of gate malfunction. Separately operable than spillways or an intentional dike opening by gate parts may also be favourable in view of dike blasting or excavation. In most cases, ad- improving inflow control, as demonstrated in justable control structures are necessary for this study. safety reasons so as to enable an interruption of Apart from optimising gate control, a good the filling process when the flood polder capac- flood forecast will also facilitate early commu- ity is reached or in unforeseen incidents. Also, nication to residents and farmers so as to allow they allow for saving storage capacity in the for loss alleviation. During a flood event, the case of a second successive flood wave and for earlier farmers are informed about the flood controlling the time of water release during the polder utilisation, the more time they have to emptying process. However, control structures undertake loss-reduction measures, such as are often uneconomic if they are only used for bringing in the harvest, evacuating livestock, or the purpose of flood protection during very removing machinery from the flood-prone area. rare flood events. Only one of the six reservoirs Reducing the amount of organic material on of the Havelpolder system is equipped with the fields by cutting the grass and bringing in control structures, whereas the others were the harvest is also advisable in view of water opened by dike blasting or excavation during quality. The more easily degradable organic the 2002 flood. A comparative analysis showed matter remains on the fields, the stronger the that the costs for construction and maintenance oxygen depletion. Particularly under warm and of the control structures exceed by far the costs calm weather conditions, oxygen concentration for dike blasting and reparation in the case of in the water may fall below levels that are lethal utilisation during rare flood events (Ellmann for fish and other aquatic and terrestrial ani- and Sauer, 2004). Dike blasting may even be mals. more economically favourable if demolition In view of water quality, it is recommended chambers are installed. The picture may change to fully utilise the flood polder storage capacity. if additional purposes of the control structures Particularly in summer, larger water depths will are included, such as frequent use for so-called dampen water warming and hence contribute to ecological floodings. higher oxygen saturation levels since cold water Ecological flooding (also called managed can hold more dissolved oxygen than warm wa- flooding) refers to the periodic flooding of low- ter. Furthermore, the possibility of allowing a lying polder areas during small flood events. It continuous flow through the flood polder allows for the development of flood-adapted should be considered so as to increase the flow wetland species and therefore reduces future velocity and hence increase oxygenation due to damage in the rare case of flood polder utilisa- flow-induced re-aeration. Generally, storage tion for flood protection. Ecological flooding times should be kept short by starting the emp- often necessitates a permanent land-use change, tying process as soon as possible so as to re- such as a conversion of intensive arable land to duce degradation of biomass and hence de- extensive grassland. This involves yearly com- oxygenation of the retained water. However, pensation transfers to the farmers or one-off during the emptying process additional control payments in case of land purchase. In the case may be required in cases where catchment man- of the Havelpolder system, opportunity costs agement objectives set constraints on the quan- for land use change would by far exceed the tity and quality of water release (Hall et al., damage reduction during rare flood events (Ell- 1993). In respect of the quantity of water re- mann and Sauer, 2004). In other riverine areas, lease, the emptying process should not start ear- such as on the River Rhine, ecological flooding lier than when safe discharges at downstream schemes, which combine objectives of flood river reaches are assured. Regarding water qual- protection and nature conservation, are already 92 Conclusions and recommendations in operation (Bettmann and Bauer, 2005; Arm- tance eventually led to an end of the plans bruster et al., 2006). (Roth and Warner, 2007). Stakeholders will ac- According to the German “Act to improve cept new proposals that affect their local envi- preventive flood control” (BMU, 2005), flood ronment more readily if they are consulted in polders are designated as flood plain areas and the early planning stages and are encouraged to each federal state is obliged to adopt land use make suggestions as to how the project could regulations in order to protect or improve ecol- be modified to meet local requirements (Hall et ogy of water bodies and their flood areas, pre- al., 1993). vent and alleviate flood damage and prevent Farmers’ acceptance is also raised by assur- erosion. The water resources laws of the federal ing financial compensation in the case of flood states of Brandenburg and Saxony-Anhalt, polder utilisation. After the flooding in 2002 where the study areas are located, prohibit con- farmers experienced yield losses due to exces- version of grassland to arable land in flood sive growth of weeds or changes in grassland plain areas. Moreover, arable land should be species assemblage. Grassland in the Havelpol- converted to grassland if possible. In the origi- der system only reached its original state after nal version of the German preventive flood four years (IaG, 2006). However, these long- control act, a strict prohibition of tillage in un- term effects are usually not compensated nor specified “flow regions” of the flood plain areas are they included in vulnerability assessments. was proposed. However, this clause was not In view of cost-efficiency, an increase in adopted in the final version of the act as the stored volume in the flood polder will hardly danger of erosion and pollution of the river was affect the expected loss in the flood polder, unproven scientifically (Munk, 2005). In fact, in whereas it may contribute considerably to a de- most of the flood plain areas sedimentation crease in the extent of affected areas, water processes dominate. Erosion is merely expected depth and hence potential loss in downstream to take place in zones and at times of high flow areas. If only a portion of the full storage capac- velocities. Therefore considerable erosion in ity of a flood polder system is required from a flood polders is mainly restricted to areas near hydraulic point of view, it may be considered the inlet gates during times of filling. Apart preferably to utilise only those flood polder res- from the construction of stilling basins, erosion ervoirs with the least damage potential and the may be reduced by filling and emptying flood lowest amount of degradable biomass at the polders at the lowest lying part if gate location time of flooding. However, hydraulic considera- is not restricted by other factors. This study tions for an optimization of the peak reduction showed that grassland in flooded areas is not effect should have first priority. If only selected generally preferable over arable land in terms of polder reservoirs are used, dike stability of the economical vulnerability and oxygen depletion. unused reservoir dikes must be ensured. In fact, there are periods such as shortly after harvest when losses on grassland outweigh crop 2.2 Future research losses. Similarly, the amount of water de- oxygenation due to surface biomass degrada- In this thesis, it was demonstrated that the tion varies considerably throughout the year analysis of flood risk reduction by means of and is subject to agricultural field operations flood polders requires comprehensive interdis- prior to flooding. Unmown grassland may con- ciplinary approaches, including methods from stitute a larger source of easily degradable bio- hydraulics, socio-economics and ecology. mass than do tilled fields. It was also stressed that the number of proc- The selection of appropriate locations for esses considered in the applied models and the flood polders often leads to conflicts with models’ spatial discretisation should always cor- farmers and land-owners, although there is a respond to the availability of observation data general agreement in the necessity of such and appropriate parameter values. While there measures. Even more opposition is expected in were few water level and water quality meas- the case of land use restrictions or ecological urement data available in the study area 1, the flooding schemes. The importance of public lack of calibration and validation data in study participation was clearly demonstrated by the area 2 made it advisable to keep model com- case of a planned flood polder designation at plexity low. Nevertheless, there is still consider- the Dutch-German border, where public resis- able improvement potential in the applied hy- Chapter VII 93 draulic and water quality models so as to better The depletion of dissolved oxygen during describe the processes that take place during long flood water storage is a serious adverse flood polder utilisation. Potential improve- ecological side effect. Consequently, it was in- ments in the hydraulic models include the investigated in this thesis and may contribute to troduction or refinement of groundwater proc- an ecological assessment of flood polder opera- esses. The water quality model could be further tion. Further environmental indicators to be in- refined by adding more processes than those cluded when assessing flood polder measures currently considered, such as de-oxygenation may be the deposition of pollutants (Wurms due to sediment oxygen demand or deriving and Westrich, 2008) or impacts on vegetation water temperature from solar radiation. growth (Armbruster et al., 2006), which were, The results of the hydraulic simulations sug- however, beyond the scope of this study. gest that the shape of a flood wave strongly in- These along with other criteria, including the fluences the potential flood peak reduction, studied flood peak reduction and economic while the probability (which in term determines losses, may contribute to an overall assessment the frequency of flood polder operation) and of flood polder measures in a multi-criteria the seasonality of flooding are of particular im- analysis (MCA). MCA techniques may either be portance when assessing costs and benefits of applied to evaluate alternative flood protection flood polder utilisation. At the same time, measures that contribute to the flood risk re- changes in flood frequency and timing due to duction of a certain area or they may be applied climate change are projected in many studies, to evaluate alternative variants of a specific e.g., Christensen and Christensen (2003). Taken flood protection measure. Alternative variants to extremes, this could mean that the discharge of a flood polder measure may refer to the de- corresponding to a 100 year flood under cur- sign and location of polder dikes and control rent conditions may be observed on average structures, operational strategies or different every 50 years or even more often. This would land use schemes. necessitate a more frequent flood polder opera- MCA methods also allow for the consideration and has to be accounted for in the design tion of non-monetary criteria in the evaluation of flood protection measures in general. There- process. After criteria evaluation is completed, fore, future research should aim at predicting weights are assigned to the specific criteria. changes in flooding probability, seasonality and Weighting is the most crucial part in a MCA. It flood wave characteristics as it determines the is often very controversial, especially when sev- effectiveness of flood control measures such as eral stakeholder groups are involved. In the in- flood polders. vestigated study areas several objectives such as In the economic analysis only direct tangible agriculture, fishery, nature conservation or rec- damage was considered. For a comprehensive reation must be taken into consideration. Cur- evaluation of flood management strategies, rently, most studies comprise social, economic though, indirect as well as intangible damage and environmental risk criteria. However, only should be taken into account. Whereas ap- few examples of real applications of MCA in proaches exist for the inclusion of indirect flood risk management exist (Meyer et al., damage like disruption to traffic, production or 2008). Further research should aim at improv- trade, an appropriate assessment of intangible ing MCA techniques for flood risk management damage such as damage to human health and to adopt them successfully in real cases involv- ecological side effects is a major challenge for ing relevant stake holders and decision makers. further research.

References 95

BfG. 2004. Deutsches Gewässerkundliches Jahrbuch. Bun- desanstalt für Gewässerkunde. Koblenz. http://dlr.bafg.de:8086/index2.php?page=jahreshaupt REFERENCES wer- te.php&K1=&K2=12845&jahr=2004&jahr_typ=abflu Alphen, J van, Marini F, Loat R, Slomp R and Passchier, R. ssjahr&MGNAME=QBI. 2008. Flood risk mapping in Europe, experiences and BfG. 2006. Wasserstands- und Abflussvorhersagen im El- best practice. Proceedings of the 4th International begebiet (Water level and discharge forecasting in the Symposium on Flood Defence, 6-8 May 2008, Toronto, Elbe basin, in German). Bundesanstalt für Gewässer- Canada, 150-1 – 150-8. kunde. Koblenz. Ambrose R B J, Wool T A Martin J L, Shanahan P and BMU. 2005. Act to improve preventive flood control. Alam M M. 2001. WASP - Water quality analysis simu- German Federal Ministry for the Environment. lation program, Version 5.2-MDEP, Model documen- http://www.bmu.de/files/pdfs/allgemein/application tation, Environmental Research Laboratory & AScI /pdf/hochwasserschutzgesetz_en.pdf. Corporation, Athens, Georgia. Böhme M et al. (eds.). 2005. Schadstoffbelastung nach dem Antenucci J and Imerito A. 2002. Dynamic reservoir simu- Elbe-Hochwasser 2002 (Contaminant loads after the lation model (DYRESM), Science Manual, Centre for Elbe flood in 2002, in German). 101 p. water research, University of Western Australia. Bowie G L, Mills W B, Porcella D B, Campbell C L, Pa- Armbruster et al. (eds.). 2006. FOWARA. Forested Water genkopf J R, Rupp G L, Johnson K M, Chan P W H, Retention Areas. Guidelines for decision makers, forest Gherini S A and Chamberlin C E. 1985. Rates, con- managers and land owners. http://www.landespflege- stants and kinetics formulations in surface water quality freiburg.de/ressourcen/fowara_guideline.pdf. 84 p. modeling (2nd ed.), Report EPA/600/3-85/040, U.S. Aureli F, Maranzoni A, Mignosa P and Ziveri C. 2005. Environmental protection agency.Baptist M J, Haas- Flood hazard mapping by means of fully-2D and quasi- noot M, Cornelissen P, Icke J, van der Wedden G, de 2D numerical modeling: a case study. In: Van Alphen J, Vriend H and Gugic G. 2006. Flood detention, nature van Beek E and Taal M (eds.). 2005. Floods, from De- development and water quality in a detention area fence to Management. Taylor & Francis. London. along the lowland river Sava, Coratia. Hydrobiologia Banks R B and Herrera F F. 1977. Effect of wind and rain 565: 243-257. on surface reaeration, J. Environ. Eng. Div. Am. Soc. Bronstert A. (ed.) 2004. Möglichkeiten zur Minderung des Civ. Eng., 103: 489–504. Hochwasserrisikos durch Nutzung von Flutpoldern an Bates P D and De Roo A P J. 2000. A simple raster-based Havel und Oder (Reducing flood risk by means of model for flood inundation simulation. Journal of Hy- flood polders at Havel and Odra Rivers, in German). drology 236(1-2): 54-77. Brandenburgische Umweltberichte 15. Potsdam. Bates P D, Horrit M S, Hunter N M, Mason D and Cobby http://pub.ub.uni- D. 2005. Numerical modelling of floodplain flow. In: potsdam.de/zsr/bub/door/door15.htm. p. 212. Bates P D, Lane S N and Ferguson R I (eds.). 2005. Büchele B, Mikovec R, Ihringer J and Friedrich F. 2004. Computational Fluid Dynamics. Applications in Envi- Analysis of potential flood retention measures (Polders) ronmental Hydraulics. John Wiley & Sons. Chichester. at the Elbe River in Saxony-Anhalt. In: Geller W et al. Bates P D, Lane S N and Ferguson R I. 2005. Computa- (Eds.). 2004. 11th Magdeburg Seminar on Waters in tional Fluid Dynamics: Applications in Environmental Central and Eastern Europe: Assessment, Protection, Hydraulics. John Wiley & Sons Ltd. Management. Proceedings of the international confer- Bettmann T and Bauer H. 2005. Hochwasserschutz am ence 18-22 October 2004 at the UFZ. Rhein durch den Polder Ingelheim (Flood protection at Busch N and Hammer M. 2007. Auswirkungen von ge- the River Rhine by the flood polder Ingelheim, in Ger- planten Rückhaltemaßnahmen an der Elbe in Sachsen man). Wasser und Abfall 9: 41-45. und Sachsen-Anhalt auf Hochwasser der Elbe (Effects BfG. 2001. Deutsches Gewässerkundliches Jahrbuch. Bun- of planned retention measures on floods at the Elbe desanstalt für Gewässerkunde. Koblenz. River in Saxony and Saxony-Anhalt, in German). http://dlr.bafg.de:8086/DGJ/PDF/2001/QBI/10046 Dresdner Wasserbauliche Mitteilungen. Heft 35. p. 003.p.pdf. 485-494. BfG. 2002. Das Augusthochwasser 2002 im Elbegebiet Chapra S C. 1997. Surface water quality modeling, (The August 2002 flood in the Elbe basin, in German). McGraw-Hill. Bundesanstalt für Gewässerkunde. Koblenz. Chatterjee C, Förster S and Bronstert A. 2008. Comparison http://elise.bafg.de/servlet/is/3967/. p. 49. of Hydrodynamic Models of Different Complexities to BfG. 2002a. Deutsches Gewässerkundliches Jahrbuch. Model Floods with Emergency Storage Areas. Hydro- Bundesanstalt für Gewässerkunde. Koblenz. logical Processes. Published online, DOI: http://dlr.bafg.de:8086/DGJ/PDF/2002/QBI/50235 10.1002/hyp.7079. 0.p.pdf. Chow V T. 1959. Open-channel hydraulics. McGraw-Hill BfG. 2002b. Deutsches Gewässerkundliches Jahrbuch. Book Company. New York. Bundesanstalt für Gewässerkunde. Koblenz. Christensen J H and Christensen O B. 2003. Severe sum- http://dlr.bafg.de:8086/DGJ/PDF/2002/QBI/58079 mertime flooding in Europe. Nature, 421: 805. 0.p.pdf. Citeau J-M. 2003. A New Control Concept in the Oise BfG. 2003. Deutsches Gewässerkundliches Jahrbuch. Bun- Catchment Area: Definition and Assessment of Flood desanstalt für Gewässerkunde. Koblenz. Compatible Agricultural Activities, FIG working week, http://dlr.bafg.de:8086/DGJ/PDF/2003/QBI/23700 Paris, France. 205.p.pdf. 96 References

Consuegra D, Joerin F and Vitalini F. 1995. Flood Delinea- Förster S, Chatterjee C and Bronstert A. 2008. Hydrody- tion and Impact Assessment in Agricultural Land using namic simulation of the operational management of a GIS Technology, 177-198. In: Carrara, A and Guzzetti, proposed flood emergency storage area at the Middle F (eds.): Geographical Information Systems in Assess- Elbe River. River Research and Applications, Wiley In- ing Natural Hazards, Kluwer Academic Publishers. terScience. Published online, DOI: 10.1002/rra.1090. DHI. 1997. MIKE11 GIS Reference and User Manual. Förster S, Kneis D, Gocht M and Bronstert A. 2005. Flood Danish Hydraulic Institute: Horsholm. Denmark. risk reduction by the use of retention areas at the Elbe DHI. 2000. MIKE 21 User Guide. Danish Hydraulic Insti- River. Intl. J. River Basin Management 3(1): 21-29. tute, Horsholm. Denmark. Fread D L, Lewis J M, Carroll T R, Rost A A and Ingram J. DHI. 2003. ECO Lab Reference manual, Danish Hydraulic 1998. Improving real-time hydrologic services in USA, Institute. Part II: Inundation mapping using dynamic streamflow DHI. 2004. MIKEFLOOD 1D-2D modelling package, modeling. British Hydrological Society International Danish Hydraulic Institute. Symposium on Hydrology in a Changing Environment. Dhondia Z and Stelling G. 2002. Application of one- UK. dimensional - two-dimensional integrated hydraulic Galbáts Z. 2006. Flood Control Management with special model for flood simulation and damage assessment. In- Reference to Emergency Reservoirs. In: Marsalek J, ternational Conference on Hydroinformatics, UK, Vol. Stancalie G and Balint G. 2006. Transboundary Floods: 5: 265-276. Reducing Risks through Flood Management. NATO Dijkman J P M, Ogink H J M, Klijn F and Van der Most Science Series IV: Earth and Environmental Sciences. H. 2003. Toelichting aanvullend deskundigenoordeel Kluwer Academic Publishers. Springer. Dordrecht, The noodoverloopgebieden. Report Q3570. WL | Delft Netherlands. Hydraulics, Delft, The Netherlands (In Dutch). Gocht M and Bronstert A. 2006. Kosteneffizienz von Flu- DKKV. 2004. Flood Risk Reduction in Germany. Lessons tungspoldern (Cost-effciency of flood polders, in Ger- Learned from the 2002 Disaster in the Elbe Region. man). In: Jüpner R. (ed.). Beiträge zur Konferenz „Stra- Summary of the study. German Committee for Disas- tegien und Instrumente zur Verbesserung des ter Reduction. vorbeugenden Hochwasserschutzes“. Internationale Dutta D, Srikantha H and Katumi M. 2003. A mathemati- Konferenz, 23-25 November 2006, Tangermünde, cal model for flood loss estimation, Journal of Hydrol- Germany. Schriftenreihe des Instituts für Wasserwirt- ogy 277, 24-49, 2003. schaft und Ökotechnologie der Universität Magdeburg- DVWK. 1985. Ökonomische Bewertung von Hochwasser- Stendal, Band 6. Shaker Verlag, Aachen: 173-182. schutzwirkungen (Economical assessment of flood pro- Hall M J, Hockin D L and Ellis J B. 1993. Design of flood tection measures, in German). Deutscher Verband für storage reservoirs. Construction Industry Research and Wasserwirtschaft und Kulturbau, DVWK Mitteilungen Information Association, CIRIA. London. 10. Hammersmark C T. 2002. Hydrodynamic modelling and Ellmann H and Schulze B. 2004. Schadenpotentiale in GIS analysis of the habitat potential and flood control Siedlung/Infrastruktur und Ökologie (Damage potenti- benefits of the restoration of a leveed Delta Island, al in settlements and infrastructure, in German). In: M.Sc. thesis, Hydrological Sciences, University of Cali- Bronstert, A (ed.). Möglichkeiten zur Minderung des fornia. Davis. Hochwasserrisikos durch Nutzung von Flutpoldern an Helms M, Büchele B, Merkel U and Ihringer J. 2002. Statis- Havel und Oder, Brandenburgische Umweltberichte tical analysis of the flood situation and assessment of 15. Potsdam. the impact of diking measures along the Elbe (Labe) Ellmann H and Sauer W. 2004. Kostenbilanzierung Bauva- river. Journal of Hydrology 267: 94–114. rianten und Sprengung (Comparison of costs between Hess T and Morris J. 1988. Estimating the Value of Flood construction variants and dike destruction, in German). Alleviation on Agricultural Grassland, Agricultural Wa- In: Bronstert, A (ed.). Möglichkeiten zur Minderung ter Management 15: 141-153. des Hochwasserrisikos durch Nutzung von Flutpoldern Hesselink A W, Stelling G S, Kwadijk J C J and Mid- an Havel und Oder, Brandenburgische Umweltberichte delkoop H. 2003. Inundation of a Dutch river polder, 15. Potsdam. sensitivity analysis of a physically based inundation EU. 2007. Directive 2007/60/EC of the European Parlia- model using historic data. Water Resources Research ment and of the Council of 23 October 2007 on the as- 39(9): 1234. sessment and management of flood risks. Official Jour- Hindmarsh A. 1983. ODEPACK, A Systematized Collec- nal of the European Union 6 November 2007. L tion of ODE Solvers, in: Stepleman, R.W. et al. (Ed.) 288/27-34. Scientific computing, pages 55–64, North-Holland, Faganello E and Attewill L. 2005. Flood Management Amsterdam. Strategy for the Upper and Middle Odra River Basin: Hoes O and Schuurmans W. 2005. Flood Standards or Feasibility Study of Raciborz Reservoir. Natural Ha- Risk Analyses for Polder Management in the Nether- zards 36: 273-295. lands, ICID 21st European Regional Conference, FAS. 1975. Modellversuche für Flutungsbauwerke (Model Frankfurt (Oder) and Slubice, Germany and Poland. experiments for flood control structures, in German), Horritt M S. 2005. Parameterisation, validation and uncer- Abschlussbericht der Forschungsanstalt für Schifffahrt, tainty analysis of CFD models of fluvial and flood hy- Wasser- und Grundbau. Berlin. draulics in the natural environment. In Computational Fischer M, Schindler M and Strobl T. 2005. Controlled Fluid Dynamics: Applications in Environmental Hy- Flood Polders - an effective method for reducing draulics, Bates P D, Lane S N and Ferguson R I (eds). floods in middle reaches of rivers. Proceedings of John Wiley & Sons Ltd.. 193-213. XIIth World Water Congress in New Delhi. References 97

Howitt J, Baldwin D, Rees G and Williams J. 2007. Model- Naturschutz und Landschaftspflege in Brandenburg 12 ling blackwater: Predicting water quality during flood- (3): 92-94. ing of lowland river forests, Ecological Modelling 203: Kotz S and Nadarajah S. 2000. Extreme value distributions 229-242. - theory and applications; Imperial College Press. Horritt M S and Bates P D. 2001. Predicting floodplain in- KTBL. 2006. Betriebsplanung Landwirtschaft 2006/07 undation: raster-based modelling versus the finite- (Agricultural engineering, in German), Kuratorium für element approach. Hydrological Processes 15: 825-842. Technik und Bauwesen in der Landwirtschaft. Darm- Horritt M S and Bates P D. 2002. Evaluation of 1D and stadt. 2D numerical models for predicting river flood inunda- KTBL. 2007. Standarddeckungsbeiträge (Standard profit tion. Journal of Hydrology 268: 87-99. margins, in German). Online version, Kuratorium für Hosking J R M and Wallis J R. 1997. Regional frequency Technik und Bauwesen in der Landwirtschaft. Darm- analysis: an approach based on L-moments. Cambridge stadt. http://www.KTBL.de. University Press: Cambridge, U.K. Kundzewicz Z W. 2008. River floods in the changing cli- Huang S, Rauberg H, Apel H, Disse, M and Linden- mate – Observations and projections. Proceedings of schmidt K-E. 2007. The effectiveness of polder sys- the 4th International Symposium on Flood Defence, 6- tems on peak discharge capping of floods along the 8 May 2008, Toronto, Canada, 141-1 – 141-9. middle reaches of the Elbe River in Germany. Hydrol- Kundzewicz Z, Ulbrich U, Brücher T et al. 2005. Summer ogy and Earth System Sciences 11: 1391-1401. Floods in Central Europe – Climate Change Track?, Hydrologic Engineering Center (HEC). 2002. HEC-RAS Natural Hazards 36: 165- 189. River Analysis System, User's Manual, Version 3.1. U.S. Kúzniar P, Popek Z and Zelazo J. 2002. The analysis of Army Corps of Engineers. Davis. California. possibility of flood risk decreasing on the Middle Vis- IaG. 2006. Gemeinsames Gutachten der Länder Branden- tula River reach. 5th International Conference on Hy- burg und Sachsen-Anhalt zur Flutung der Havelniede- dro -Science & -Engineering, Warsaw. rung bei Hochwasserereignissen. Ökologische Aspekte http://kfki.baw.de/conferences/ICHE/2002- der Flutung. (Report of the federal states of Branden- Warsaw/ARTICLES/PDF/185C2.pdf burg and Saxony-Anhalt on the flooding of the Havel LAWA. 1995. Guidelines for Forward-Looking Flood Pro- River lowland. Ecological aspects of the flooding. In tection. Länderarbeitsgemeinschaft Wasser. German). Institut für angewandte Gewässerökologie. Lecce S. 2000. Seasonality of Flooding in North Carolina, Unpublished report. Southeastern Geographer 21, 2: 168-175. IKSE. 2003. Aktionsplan Hochwasserschutz Elbe (Elbe LfL. 2005. Veränderte Landnutzungssysteme in hochwas- flood action plan, in German). International Commis- sergefährdeten Gebieten (Land use change in flood sion for the Protection of the Elbe River. prone areas, in German). Schriftenreihe der Sächsi- http://elise.bafg.de/servlet/is/5130/. p. 79. schen Landesanstalt für Landwirtschaft, Heft 12. IKSE. 2004. Dokumentation des Hochwassers vom Au- LfUG. 2005. Hochwasser in Sachsen – Gefahrenhinweise- gust 2002 im Einzugsgebiet der Elbe (Documentation karte (Floods in Saxony – Hazard maps, in Geman). of the August 2002 flood at the Elbe River ,in Ger- Sächsisches Landesamt für Umwelt und Geologie. man). International Commission for the Protection of LfUG. 2006. Sächsisches Landesamt für Umwelt und Geo- the Elbe River. http://elise.bafg.de/servlet/is/6889/. logie. Gewässerkundliche Statistik. p. 207. http://www.umwelt.sachsen.de/de/wu/umwelt/lfug/l IKSE. 2005. Die Elbe und ihr Einzugsgebiet (The Elbe Ri- fug-internet/documents/560051_Q2006.pdf. ver and its catchment, in German). International Com- LHW. 2003. Deutsches Gewässerkundliches Jahrbuch. El- mission for the Protection of the Elbe. 258 p. begebiet. Teil I. Landesbetrieb für Hochwasserschutz IWK. 2003. Untersuchung von Hochwasserretentions- und Wasserwirtschaft Sachsen-Anhalt. maßnahmen entlang der Elbe im Bereich der Landkrei- Lindenschmidt K-E. 2006. The effect of complexity on pa- se Wittenberg und Anhalt-Zerbst (Study on flood re- rameter sensitivity and model uncertainty in river water tention measures along the Elbe River in the counties quality modelling, Ecological Modelling, 190: 72–86. Wittenberg and Anhalt-Zerbst, in German). Institut für Lindenschmidt K-E, Huang S and Baborowski M. 2008. A Wasserwirtschaft und Kulturtechnik. Universität Karls- quasi-2D flood modeling approach to simulate sub- ruhe. stance transport in polder systems for environment Jaffe D A and Sanders B F. 2001. Engineered Levee Brea- flood risk assessment, Science of the total environment, ches for Flood Mitigation. Journal of Hydraulic Engi- 397(1-3): 86–102. neering 127(6): 471-479. McCutcheon S C. 1989. Water quality modeling, Vol. 1 - Johnson N L, Kotz S and Balakrishnan N. 1994. Continu- Transport and surface exchange in rivers, CRC press. ous Univariate Distributions, 2nd ed., Wiley- MEDIS. 2007. Methods for the evaluation of direct and Interscience. indirect losses. In: RIMAX-Risk Management of Ex- King I P, Donnell B P, Letter J V, McAnally W H, Thomas treme Flood Events. http://www.rimax- W A and LaHatte C. 2001. User's Guide for RMA2 ho- Version 4.5. chwasser.de/fileadmin/RIMAX/download/Allgemeine Knight D W and Shiono K. 1996. River Channel and s/rimax_broschuere_auflage2.pdf. Floodplain Hydraulics. In: Anderson, M G, Walling, Merz B, Kreibich H, Thieken A and Schmidtke R. 2004. des E and Bates P D (eds.). Floodplain Processes. John Estimation uncertainty of direct monetary flood dam- Wiley & Sons Ltd. age to buildings, Natural Hazards and Earth System Knösche R. 2003. Fischökologische und fischereiliche Science 4: 153-163. Schäden durch Extemhochwasser (Damages to fish e- cology and fishery due to extreme floods, in German). 98 References

Merz B and Thieken A. 2005. Separating natural and epis- east Germany, Hydrology and Earth System Sciences temic uncertainty in flood frequency analysis, Journal 11: 1-14. of Hydrology 309: 114-132. Petzold L. 1983. Automatic Selection of Methods for Solv- Merz B, Thieken A and Gocht M. 2005. Flood Risk Map- ing Stiff and Nonstiff Systems of Ordinary Differential ping at the Local Scale: Concepts and Challenges. In: Equations, Siam J. Sci. Stat. Comput., 4: 136–148. Flooding in Europe: Challenges, Developments in Peukert V. 1970. Untersuchungen über den Einfluss von Flood Risk Management. Kluwer. überstauten Flächen auf die Wasserqualität von Tal- Merz B Thieken A and Gocht M. 2007. Flood risk map- sperren (Investigations on the impact of inundated land ping at the local scale: concepts and challenges. In: on the water quality of reservoirs, in German), Fort- Flood Risk Management in Europe, Advances in Natu- schritte der Wasserchemie und ihrer Grenzgebiete, 12: ral and Technological Hazards Research 25, Springer, 66–82. 231 – 251. Pivot J-M, Josien E and Martin P. 2002. Farm adaptation Messner F, Penning-Rowsell E, Green, C et al. 2007. to changes in flood risk: a management approach, Jour- Evaluating flood damages: guidance and recommenda- nal of Hydrology 267: 12-25. tions on principles and methods. Plate E. 2002. Flood risk and flood management. Journal http://www.floodsite.net/html/partner_area/project_ of Hydrology. 267: 2-11. docs/T09_06_01_Flood_damage_guidelines_D9_1_v2 R Development Core Team. 2007. R: A Language and En- _2_p44.pdf. vironment for Statistical Computing, R Foundation for Meyer V and Messner F. 2005. National Flood Damage Statistical Computing, Vienna, Austria, ISBN 3-900051- Evaluation Methods, A Review of Applied Methods in 07-0, URL http://www.R-project.org. England, the Netherlands, the Czech Republic and Rátky I and Szlávik L. 2001. Perfection of Operation Con- Germany. UFZ-Discussion Papers 21. trol for the Emergency Reservoirs in the Körös Valley. Meyer V, Scheuer S and Haase D. 2008. A multicriteria ap- Periodica Polytechnica Ser. Civ. Eng. 45(2): 93-119. proach for flood risk mapping exemplified at the Mulde Reichert P, Borchardt D, Henze M, Rauch W, Shanahan P, river, Germany. Natural Hazards. Somlyody L and Vanrolleghem P A. 2001. River water Minh Thu P T. 2002. A hydrodynamic-numeric model of quality model No. 1, IWA Publishing. the river Rhine. Dissertation. University of Karlsruhe. Romero J, Hipsey M, Antenucci J and Hamilton D. 2004. MLUR (Ed.). 2007. Nutzung von Biomasse als Gärsub- Computational Aquatic Ecosystem Dynamics Model strate (Use of biomass as a substrate for fermentation, (CAEDYM v.2.1), Science Manual, Centre for water in German). 26. Fachtagung Acker- und Pflanzenbau. research, University of Western Australia. Landesamt für Verbraucherschutz, Landwirtschaft und Roth D and Warner J. 2007. Flood risk, uncertainty and Flurneuordnung Brandenburg. changing river protection policy in the Netherlands: Munk H-H. 2005. Das Gesetz zur Verbesserung des vor- The case of “calamity polders”. Tijdschrift voor Eco- beugenden Hochwasserschutzes (The act to improve nomische en Sociale Geografie 98, 4: 519-525. preventive flood control, in German). Wasser und Ab- Rungo M and Olesen K W. 2003. Combined 1- and 2- di- fall 5: 20-23. mensional flood modeling. 4th Iranian Hydraulic Con- MURL. 2000. Hochwasserschadenpotentiale am Rhein in ference, 21-23 Oct, Shiraz, Iran. NRW (Flood damage potential at the River Rhine in Schmidt M. 1957. Gerinnehydraulik (Channel hydraulics, in North Rhine-Westfalia, in German). Ministerium für German). VEB Verlag Technik Berlin und Bauverlag Umwelt, Raumordnung und Landwirtschaft des Landes GmbH. Wiesbaden. Nordrhein-Westfalen. Unpublished. Shanahan P, Henze M, Koncsos L, Rauch W, Reichert P, Neubert G and Thiel R. 2004. Schadenpotentiale in der Somlyody L and Vanrolleghem P. 1998. River Quality Landwirtschaft (Damage potential in agriculture, in Modelling: II. Problems of the Art. IAWY Biennial In- German). In: Bronstert, A. (ed.) Möglichkeiten zur ternational Conference, Vancover, British Columbia, Minderung des Hochwasserrisikos durch Nutzung von Canada, 21-26 June 1998. Flutpoldern an Havel und Oder, Brandenburgische Silva W, Dijkman J and Loucks D. 2004. Flood manage- Umweltberichte 15. Potsdam. ment options for the Netherlands. International Journal O’Brien J S. 2006. FLO-2D: Users Manual Version of River Basin Management 2(2): 101-112. 2006.01. http://www.flo- Sivapalan M, Blöschl G, Merz R and Gutknecht D. 2005. 2d.com/v2006/Documentation/Manual_Main_2006.p Linking flood frequency to long-term water balance: df incorporating effects of seasonality, Water Resources OECD. 2002. Evaluation and Aid Effectiveness - Glossary Research 41. doi:10.1029/2004WR003439. of Key Terms in Evaluation and Results Based Man- Smith K and Ward R. 1998. Floods: Physical Processes and agement. Organisation for Economic Co-operation and Human Impact. John Wiley & Sons. Chichester. Development. Stelling G and Duinmeijer S. 2003. A staggered conserva- http://www.oecd.org/dataoecd/29/21/2754804.pdf. tive scheme for every froude number in rapidly varied 37p. shallow water flows. International Journal for Numeri- Penning-Rowsell E, Johnson C, Tunstall S et al. 2003. The cal Methods in Fluids 43: 1329-1354. benefits of flood and coastal defence: techniques and Tayefi V, Lane S N, Hardy R J and Yu D. 2007. A com- data for 2003, Flood Hazard Research Centre, Middle- parison of one- and two-dimensional approaches to sex University. modeling flood inundation over complex upland flood- Petrow T, Merz B, Lindenschmidt K-E and Thieken A. plains. Hydrological Processes 21 (23): 3190-3202. 2007. Aspects of seasonality and flood generating circu- Todorovic P and Woolhiser D. 1972. On the Time When lation patterns in a mountainous catchment in south- the Extreme Flood Occurs. Water Resources Research 8, 6: 1433-1438. References 99

USDA. 1978. Economics – Basic Data for Evaluating Wolter C, Arlinghaus R, Grosch U A and Vilcinskas A. Floodwater Damages to Crops and Pastures in the 2003. Fische & Fischerei in Berlin (Fishes and fishery in Northeast, Technical Note, United States Department Berlin, in German). Zeitschrift für Fischkunde, Suppl. of Agriculture. 2: 1-156. Werner M. 2001. Impact of grid size in GIS based flood Wurms S and Westrich B. 2008. Targeted Retention of extent mapping using a 1-D flow model. Physics and contaminated Sediment in a green Flood Retention Chemistry of the Earth, Part B: Hydrology, Oceans and Reservoir. Proceedings of the 4th International Sympo- Atmosphere 26: 517-522. sium on Flood Defence, 6-8 May 2008, Toronto, Can- Werner M. 2004. Spatial flood extent modeling – A per- ada, 140-1 – 140-8. formance based comparison. Ph.D thesis, Delft Uni- YRFCMP. 2003. Literature Review for the Development versity, Netherlands. of a Socio-Economic Impacts Assessment Procedure Werner M, Hunter N M and Bates P D. 2005. Identifiabil- to be applied to the flooding of Qianliang Hu Deten- ity of distributed floodplain roughness values in flood tion Basin, Hunan Province, China, Yangtze River extent estimation. Journal of Hydrology 314: 139-157. Flood Control and Management Project.

101

ACKNOWLEDGEMENTS

After the large flood in Central Europe in August 2002 various research projects on flood risk management were initiated. I was lucky to get involved in two interesting projects. In the first project a small enthusiastic interdisciplinary group took up the challenge of producing reasonable results in a very short time. I would like to thank all team colleagues for their engagement, particularly David Kneis and Martin Gocht. The second project gave me the opportunity to gain insights into how a large successful EU project is coordinated, get to know the leading flood researchers in Europe and be a small part of the large FLOODsite family. I would like to thank all those who supported me during my PhD time. First of all, thanks to Axel Bronstert for giving me the opportunity to work on an interesting and up-to-date topic and letting me have all my freedom in doing so. I am also grateful to the other members of the defence committee, in particular Bruno Merz and Robert Jüpner for taking the time to read this thesis. I will keep many valuable memories of the PhD time that I don’t want to miss, like the exceptional experience of teaching twenty Chinese in Budapest about German flood risk management or of carrying heavy GPS instruments in suffocating heat for many hours to improve an out-dated polder DEM a little bit. During his one-year stay at our institute I closely co-operated with Chandranath Chatterjee. His thoughtfulness, reliability and heartiness made this time inspiring and enjoyable, both scientifically and personally. Publishing papers can be a long and at times exhausting process. I would especially like to thank Chandranath Chatterjee, David Kneis and Karl-Erich Lindenschmidt for their encouraged work to bring this process to success. I would like to thank our group at the Institute of Geoecology for creating an enjoyable atmosphere at work and on many other occasions. I also thank Annegret Thieken, Heidi Kreibich and Theresia Petrow and others from the Engineering Hydrology group at GFZ Potsdam for interesting discussions. Special thanks to Sibylle Itzerott for her continuing support, the company on the field trip and nice chats while inline skating. I am thankful to the Danish Hydraulic Institute (DHI) for providing an evaluation copy of the MIKE software and the German DHI group for giving support when I had urgent questions. I am most grateful to my family for their love and support and in particular to my parents for many hours of babysitting. Finally, I would like to thank Micha for being a wonderful husband and father, encouraged scientist and lecturer, patient listener, accompanying person, friend and lots more.

102

103

CURRICULUM VITAE AND LIST OF PUBLICATIONS

Saskia Förster was born on 5th November 1976 in Potsdam. After completing secondary school in 1996, she studied Geoecology at the University of Potsdam including a two semester stay at the Univer- sity of Southampton, UK. During her studies she specialised in GIS and remote sensing techniques. Af- ter graduating, she started her PhD work at the Institute of Geoecology at the University of Potsdam in 2003. She was involved in two research projects on flood risk management, viz., “Möglichkeiten zur Minderung des Hochwasserrisikos durch Nutzung von Flutpoldern an Havel und Oder” funded by the German Ministry of Education and Research and “Integrated Flood Risk Analysis and Management Methodologies (FLOODsite)” funded under the Sixth Framework Program of the European Commis- sion. Apart from the project and PhD work she gave GIS courses and supervised students in their study projects.

Reviewed journal articles

Kneis D, Förster S and Bronstert A. Simulating water quality in a flood detention area using models of different spatial discretisation. Ecological Modelling. (in preparation)

Förster S, Kuhlmann B, Lindenschmidt K-E and Bronstert A. 2008. Assessing flood risk for a rural detention area. Nat. Hazards Earth Syst. Sci., 8, 311–322.

Chatterjee C, Förster S and Bronstert A. 2008. Comparison of Hydrodynamic Models of Different Complexities to Model Floods with Emergency Storage Areas. Hydrological Processes. Published online, DOI: 10.1002/hyp.7079.

Förster S, Chatterjee C and Bronstert A. 2008.Hydrodynamic Simulation of the Operational Manage- ment of a Proposed Flood Emergency Storage Area at the Middle Elbe River. River Research and Applications. Published online, DOI: 10.1002/rra.1090.

Förster S, Kneis D, Gocht M and Bronstert A. 2005. Flood Risk Reduction by the Use of Retention Areas at the Elbe River. Intl. Journal of River Basin Management, 3 (1), 21-29.

Non-reviewed articles

Förster S, Chatterjee C and Bronstert A. 2007. Investigation on the Operation of a Proposed Emer- gency Flood Storage Area at the Middle Elbe River in the Context of the Integrated Project FLOODsite. ERWG Letter, Land and Water Management in Europe. 17, 4-5.

Conference proceedings and papers

Förster S and Bronstert A. 2008. Assessment of hydraulic, economic and ecological impacts of flood polder management – A case study from the Elbe River, Germany. Floodrisk2008. 30 September – 2 October 2008, Oxford, UK.

Förster S, Kneis D and Bronstert A. 2008. Sauerstoffhaushalt im Flutungspolder. Tag der Hydrologie, 27. - 28. März 2008, Hannover, Germany.

Förster S, Chatterjee C and Bronstert A. 2007. Reducing flood risk by emergency storage - A case study from the Elbe River. In: Schanze J (Ed.). 2007. Flood Risk Management Research. From ex- 104

treme events to citizens involvement. Proceedings of the European Symposium on Flood Risk Man- agement Research, 6 - 7 February 2007, Dresden, Germany, 217-219.

Förster S, Chatterjee C and Bronstert A. 2006. Hochwasserüberschwemmungssimulation mittels hyd- rodynamischer Modellierung für das Management eines geplanten Polderstandortes an der Mittleren Elbe. In: Jüpner R (Ed.). Beiträge zur Konferenz „Strategien und Instrumente zur Verbesserung des vorbeugenden Hochwasserschutzes“. Internationale Konferenz, 23. - 25. November 2006, Tanger- münde, Germany. Schriftenreihe des Instituts für Wasserwirtschaft und Ökotechnologie der Univer- sität Magdeburg-Stendal, Band 6. Shaker Verlag, Aachen, 83-92.

Förster S, Chatterjee C, De Medina V, Bateman A and Bronstert A. 2006. Flood Inundation Simula- tion using Hydrodynamic Models of Different Complexity for the Management of a Potential Pol- der at the Middle Elbe River, Germany. 3rd International Symposium on Integrated Water Re- sources Management, 26 – 28 September 2006, Bochum, Germany.

Förster S and Bronstert A. 2005. Benefits and Costs of Using Detention Areas for Flood Risk Reduc- tion at the Elbe River. In: van Alphen J, van Beek E and Tall M (Eds.). 2005. Floods from Defence to Management. 3rd International Symposium on Flood Defence, 25 - 27 May 2005, Nijmegen, The Netherlands, 285.

Förster S, Kneis D, Gocht M and Bronstert A. 2004. Flood Risk Reduction by Use of Detention areas at the Elbe River. EGU General Assembly, 25 – 30 April 2004, Nice, France.

Förster S, Kneis D and Bronstert A. 2004. Untersuchung einer gesteuerten Flutung der Unteren Ha- velniederung zur Minderung des Hochwasserrisikos an der Elbe. Tag der Hydrologie, 22. - 23. März 2004, Potsdam, Germany. 105

AUTHOR’S DECLARATION

I prepared this dissertation without illegal assistance. The work is original except where indicated by special reference in the text and no part of the dissertation has been submitted for any other degree. This dissertation has not been presented to any other University for examination, neither in Germany nor in another country.

Saskia Förster Potsdam, August 2008